# 黄仁勋 · Jensen Huang — 封面传记 ACF-00-00160

> 亚洲封面人物 Asia Cover Figure · 机器可读档案（LLM-ready）。本文件由官网结构化档案数据自动生成，供 AI 检索与引用。中文全文与英文全文对照编排。

## 档案元数据 Metadata

- 封面编码 ACF Code：**ACF-00-00160**
- 姓名 Name：黄仁勋 / Jensen Huang
- 职务 Title：联合创始人、总裁兼首席执行官 / Co-founder, President and CEO
- 公司 Company：英伟达（NVIDIA） / NVIDIA Corporation
- 出生 Born：1963-02-17，中国台湾台北市（在台南市长大） (Taipei, Taiwan (raised in Tainan))
- 篇别 Category：格局（格局篇 / Cover Biography (Geju)）
- 入档日期 Accessioned：2026-07-20
- 标签 Tags：半导体, GPU, 人工智能, 英伟达, CUDA, 芯片, AI基础设施
- 永久档案链接 Archive URL：https://coverfigure.com/acf/ACF-00-00160/geju
- English archive：https://coverfigure.com/acf/ACF-00-00160/geju?lang=en
- 官网原文报道 Feature story：https://coverfigure.com/acf/figure/jensen-huang

## 信息口径声明 Editorial Sourcing Policy

- 工商登记信息可通过国家企业信用信息公示系统核验；经营数据来自企业公开披露、官方发布与权威媒体报道，引用处标注来源。
- 收购、IPO、估值、融资等事项以官方公告为准；人物观点与言论为公开演讲、采访的事实记录，不代表编辑部立场。
- 本档案仅供行业研究参考，不构成投资建议。事实更正联系：editorial@coverfigure.com。

## 卷首 Editorial Intro

1973年的美国肯塔基州阿巴拉契亚山区，一个10岁的华裔男孩站在一座破旧宿舍楼的走廊里。他身材矮小，留着长发，说话带着很重的口音——他是这所学校里唯一的亚洲面孔。每天，其他男孩会在他走过人行桥时猛摇绳索，试图把他晃下去。他们叫他"Chinks"——一个对华人的侮辱性称呼。他的室友是一个满身纹身和刀疤的17岁少年，刚出狱的问题生。黄仁勋教这个文盲室友读书认字，室友则教他卧推[s3][s4]。他的父母为了凑学费几乎变卖了所有家当——他们以为这是一所"精英寄宿学校"，实际上它是面向问题青少年的宗教感化学校Oneida Baptist Institute。每天打扫完厕所后，这个男孩会躲在角落里读书。他后来说，在肯塔基的日子"比其他任何记忆都更清晰"[s3]。五十多年后，这个曾在肯塔基扫厕所的移民少年，缔造了人类历史上最有价值的公司之一。从台南的街巷到曼谷的国际学校，从肯塔基的宿舍到俄勒冈的工厂，从Denny's餐厅的卡座到全球最有影响力的科技舞台——这段旅程跨越了半个地球和半个世纪[s3][s4][s5]。母亲的教育方式塑造了他的底色——她自己不懂英文，却每天从韦氏字典里随机挑10个单词教两个儿子。母亲总是说"你真的非常特别"，这句话给了他"必须变得更出色"的甜蜜压力。他后来说："天下都知道我是一个无可救药的完美主义者，这完全遗传了自我父亲；而大家也知道，我每天都沉迷于当一个完美主义者，这把我妈灌输给我的！"[s3]。天文、数学、乒乓球、芯片——他的人生像一条不断加速的抛物线，起点是台南巷弄里的一台收音机，终点尚未到来。或许正如他在剑桥大学的演讲所说："当CEO是一生的牺牲。"这不是谦辞，而是一个从底层爬上来的人对"拥有"这件事最深刻的理解[s21]。

In 1973, in the Appalachian hills of Kentucky, a ten-year-old Chinese boy stood in the corridor of a dilapidated dormitory. He was short, wore his hair long and spoke with a heavy accent; he was the only Asian face in the school. Every day, as he crossed a footbridge, other boys would shake the ropes, trying to throw him off. They called him "Chinks," a slur for Chinese people. His roommate was a seventeen-year-old fresh out of jail, a problem child covered in tattoos and knife scars. Huang taught his illiterate roommate to read; the roommate taught him to bench-press[s3][s4]. His parents, scraping together tuition, sold nearly everything they owned. They had believed the place to be an "elite boarding school." It was in fact the Oneida Baptist Institute, a religious reform academy for troubled youth. After cleaning the toilets each day, the boy would hide in a corner and read. He later said his Kentucky days were "more vivid than any other memory"[s3]. More than fifty years later, that immigrant child who scrubbed toilets in Kentucky built one of the most valuable companies in human history. From the alleyways of Tainan to an international school in Bangkok, from a Kentucky dormitory to an Oregon factory floor, from a Denny's booth to the most influential stage in technology, the journey spans half a globe and half a century[s3][s4][s5]. His mother's methods shaped his bedrock. Speaking no English herself, she picked ten words at random from a Webster's dictionary each day and taught them to her two sons. She always told him, "You really are very special"—a phrase that gave him the sweet pressure of having to become more extraordinary. He later said: "Everybody knows I am a hopeless perfectionist—that I inherited entirely from my father. And everybody knows that I indulge in being a perfectionist every day—that was instilled in me by my mother!"[s3]. Astronomy, mathematics, table tennis, chips: his life traces a parabola that keeps accelerating, beginning with a radio in a Tainan alley and with no end yet in sight. Perhaps, as he put it in a speech at the University of Cambridge, "to be a CEO is a lifetime of sacrifice." That is not false modesty. It is the deepest understanding of "having" that a man who climbed up from the bottom can hold[s21].

## 人物速览 Lead

黄仁勋，英伟达（NVIDIA）联合创始人、总裁兼首席执行官，全球AI基础设施的缔造者，AI时代最具影响力的科技领袖之一。1963年2月17日出生于中国台湾台北市，在台南市长大，祖籍浙江青田[s1][s2]。9-10岁时被父母送往美国，却阴差阳错地被舅舅送进肯塔基州一所面向问题青少年的宗教寄宿学校，与满身纹身和刀疤的室友同住，每天打扫厕所，遭受种族欺凌[s3][s4]。15岁在Denny's餐厅上夜班洗碗，16岁考入俄勒冈州立大学[s5][s6]。1993年，30岁的黄仁勋与两位合伙人在加州圣何塞一家Denny's餐厅里决定创业，用约4万美元银行贷款创办了英伟达[s7][s8]。从第一款芯片NV1押错技术路线濒临破产，到世嘉谈判起死回生；从发明GPU概念开创图形计算新时代，到孤注一掷押注CUDA十年等待世界赶上来；从2012年AlexNet触发AI大爆炸，到ChatGPT引爆算力需求指数级增长——英伟达从一家4万美元起步的芯片小公司成长为市值突破5万亿美元、全球最有价值的企业之一[s9][s10][s11][s12]。CUDA生态系统拥有超500万开发者，成为全球AI训练和推理的事实标准[s12][s13]。2026年3月，他在Lex Fridman播客中宣称"我认为我们已经实现了AGI"，再次引发全球讨论[s14]。他以"AI的缔造者"身份入选《时代》周刊2025年度人物，位列《财富》全球100位最具影响力商界人士第一位[s11]。

Jensen Huang is the co-founder, president and chief executive officer of NVIDIA, the architect of the world's artificial-intelligence infrastructure and one of the most influential technology leaders of the AI age. He was born on February 17, 1963, in Taipei, Taiwan, and grew up in Tainan; his family's ancestral home is in Qingtian, Zhejiang Province[s1][s2]. At the age of nine or ten he was sent to the United States by his parents, only for his uncle to enroll him, by a grotesque misjudgment, at a religious boarding school in Kentucky for troubled teenagers, where he shared a room with tattooed, scarred delinquents, cleaned the toilets every day and endured racial bullying[s3][s4]. He washed dishes on the night shift at a Denny's restaurant at fifteen and was admitted to university at sixteen[s5][s6]. In 1993, at thirty, Huang and two co-founders resolved to start a company in a Denny's booth in San Jose, California, launching NVIDIA with roughly US$40,000 in bank loans[s7][s8]. The company's first chip, the NV1, backed the wrong technology and nearly bankrupted it; a crucial contract with Sega pulled it back from the edge. He coined the concept of the GPU and opened a new era of graphics computing, then bet the company on CUDA and waited a decade for the world to catch up. AlexNet ignited the AI explosion in 2012, and ChatGPT sent demand for compute into exponential growth in 2022. A chip startup begun with US$40,000 grew into one of the most valuable companies on earth, its market capitalisation surpassing US$5 trillion[s9][s10][s11][s12]. The CUDA ecosystem now numbers more than five million developers and has become the de facto standard for AI training and inference worldwide[s12][s13]. In March 2026, on the Lex Fridman podcast, he declared, "I think we've achieved AGI," setting off a fresh global debate[s14]. Time magazine named him a 2025 Person of the Year as "the builder of AI," and Fortune ranked him first among the world's 100 most influential business leaders[s11].

## 正文 Archive Chapters

### 1. 早年与求学：从台南到曼谷到肯塔基 / Early Years and Education: From Tainan to Bangkok to Kentucky

1963年2月17日，黄仁勋出生于中国台湾台北市，后随家人迁居台南。父亲黄兴泰是国立成功大学毕业的化学工程师，在炼油厂工作；母亲罗采秀是教师，出身台南望族罗氏家族[s1][s2]。家中说闽南语（台语），这是一个典型的中国台湾中产家庭，因父亲工作而经常搬家。母亲的教育方式极为独特：她自己不懂英文，却每天从韦氏字典里随机挑10个单词教两个儿子。这种近乎偏执的教育方式，让黄仁勋兄弟俩从小就对英语有了异乎寻常的敏感度。母亲总是对他说"你真的非常特别"，这句话在无形中给了他"必须变得更出色"的甜蜜压力。多年后黄仁勋在中国台湾的公司庆祝大会上动情地说："天下都知道我是一个无可救药的完美主义者，这完全遗传了自我父亲；而大家也知道，我每天都沉迷于当一个完美主义者，这把我妈灌输给我的！"[s3]。4岁半时，他逛夜市被利刃划伤脸部，留下了至今可见的疤痕——这成了他面部最显著的特征之一。

1968年，5岁的黄仁勋因父亲工作调动，随全家迁往泰国曼谷生活。他就读于曼谷Ruamrudee国际学校，在泰国度过了大约四年时光。在泰国的日子里，他和哥哥展现出强烈的冒险天性——曾将打火机燃料倒入游泳池并点燃，这种近乎鲁莽的好奇心，日后成为他商业决策中最鲜明的底色。父亲曾赴纽约一家空调公司接受培训，这段美国之行让他大开眼界，回来后决心送两个儿子去美国接受更好的教育[s3][s4]。20世纪60年代末70年代初，因越南战争蔓延导致泰国社会动荡加剧，加速了父母送走孩子的决定。

1973年，9-10岁的黄仁勋和哥哥被送往美国华盛顿州塔科马市的舅舅家——这段经历成为黄仁勋人生中最具传奇色彩、也最令人心酸的篇章。刚移民到华盛顿州的舅舅和舅妈，误将肯塔基州的Oneida Baptist Institute当成了一所"精英寄宿学校"。实际上，这是一所1899年创立的面向问题青少年的宗教寄宿学校（religious reform academy for troubled youth），最初为阿巴拉契亚地区的贫困儿童提供教育，后来也接收"处境困难"的青少年。为了凑学费，父母几乎变卖了所有家当——他们不知道自己把儿子送进了一个什么样的地方[s3][s4]。

10岁的黄仁勋是该校历史上年龄最小的寄宿生。他在那里的经历堪称磨难：室友是一个满身纹身和刀疤的17岁少年，刚出狱的问题生。黄仁勋教这个文盲室友读书认字，作为交换，室友教他卧推。每天他需要打扫男生宿舍的厕所——与他母亲（一位教师）的期望形成了刺目的对比。哥哥年龄大些，被分配到附近的烟草农场做体力劳动。作为"身材矮小的亚洲移民，留着长发，英语口音很重"的他，经常被欺负和殴打。每天面对种族辱骂——"Chinks"，一个对华人的侮辱性称呼——走过人行桥时其他男孩会猛摇绳索试图把他晃下去。他后来回忆："那时候没有辅导员可以倾诉。你只能坚强起来，继续前进。"这段经历塑造了他极高的"痛苦容忍度"——后来成为他管理哲学的核心概念[s3][s4][s23]。

但逆境中也有救赎。黄仁勋加入了游泳队，学会了打乒乓球，体育成为他在逆境中生存和建立自信的方式。14岁时，他登上了《体育画报》（Sports Illustrated）——一个来自中国台湾的移民少年，在肯塔基的山区学校里，凭着一颗乒乓球打进了美国主流体育媒体。他后来说，对肯塔基的记忆"比其他任何记忆都更清晰"。2019年，功成名就的他不计前嫌，向这所学校捐赠了200万美元，修建了以他名字命名的女生宿舍楼Jen-Hsun Huang Hall——这不是报复，而是感恩。那些年在肯塔基山区学到的坚韧，比任何商学院的课程都更加深刻[s1][s3]。

到肯塔基约两年后，即1975年前后，父母终于移民到美国俄勒冈州比弗顿市（Beaverton），全家团聚。黄仁勋进入阿罗哈高中（Aloha High School），学业突飞猛进，连跳两级，16岁毕业。他成为全美排名的乒乓球选手，同时是数学、计算机和科学俱乐部的活跃成员。1977年，学校购买了一台Apple II电脑——这是当时最先进的个人电脑之一。黄仁勋用BASIC语言编写了自己的贪吃蛇游戏版本，还玩Super Star Trek文字游戏。这是他第一次与计算机世界亲密接触，但此时他还不知道，这将是改变他一生的邂逅[s1]。

从15岁起（约1978-1983年），黄仁勋在当地Denny's餐厅做夜班（graveyard shift）——洗碗工、勤杂工、服务员，什么活都干。他后来说自己"洗过很多很多厕所"。从肯塔基的厕所到Denny's的厨房，他的人生似乎一直在与厕所打交道——但正是这些最底层的劳动，塑造了他日后管理数万人企业时的务实底色。Denny's"比家里安静，咖啡便宜"，可以坐很久。他对这家连锁餐厅产生了特殊感情——多年后，他会在Denny's的卡座里画出英伟达的蓝图[s5][s6]。

1979年，16岁的黄仁勋考入俄勒冈州立大学电子工程专业。选择这所学校的原因很务实——"州内学费便宜"。在大学实验室里，他结识了Lori Mills——当时电气工程系250名学生中只有3位女生，Lori是其中之一。两人是实验课搭档（lab partner），共同的课程和兴趣让两颗心慢慢靠近。他向Lori承诺"30岁时成为一家公司的CEO"——这个在旁人看来不切实际的承诺，后来一字不差地兑现了[s15][s16]。

**English:** Jensen Huang was born on February 17, 1963, in Taipei, Taiwan, and later moved with his family to Tainan. His father, Huang Hsing-tai, was a chemical engineer trained at National Cheng Kung University who worked at an oil refinery; his mother, Lo Tsai-hsiu, was a teacher from the prominent Lo family of Tainan[s1][s2]. The household spoke Hokkien, the Taiwanese dialect. It was a typical middle-class Taiwanese family that moved often for the father's work. His mother's educational method was extraordinary: speaking no English herself, she chose ten words at random from a Webster's dictionary every day and taught them to her two sons. That almost obsessive routine gave the Huang brothers an unusual sensitivity to English from childhood. She always told him, "You really are very special"—words that quietly imposed on him the sweet pressure of having to become more exceptional. Years later, at an emotional company celebration in Taiwan, Huang said: "Everybody knows I am a hopeless perfectionist—that I inherited entirely from my father. And everybody knows that I indulge in being a perfectionist every day—that was instilled in me by my mother!"[s3]. At the age of four and a half, he was cut in the face by a blade while wandering a night market, leaving a scar still visible today—one of the most distinctive features of his face.

In 1968, at five, Huang moved with the family to Bangkok, Thailand, after his father was transferred there for work. He attended Ruamrudee International School in Bangkok and spent about four years in Thailand. In Thailand, he and his older brother displayed a fierce appetite for adventure—once pouring lighter fluid into a swimming pool and setting it alight. That near-reckless curiosity would later become the most vivid strain in his business decisions. His father had trained at an air-conditioning company in New York, and the American trip opened his eyes; on his return he resolved to send both sons to the United States for a better education[s3][s4]. In the late 1960s and early 1970s, the spread of the Vietnam War brought growing social turmoil to Thailand, hastening the parents' decision to send the children away.

In 1973, Huang, then nine or ten, and his brother were sent to live with an uncle in Tacoma, Washington—the episode that became the most legendary, and the most poignant, chapter of Huang's life. The uncle and aunt, themselves recent immigrants to Washington State, mistook Kentucky's Oneida Baptist Institute for an "elite boarding school." In reality it was a religious boarding school founded in 1899 as a reform academy for troubled youth, originally educating poor children of Appalachia and later taking in teenagers in "difficult circumstances." To meet the tuition, the parents sold nearly everything they owned—never knowing what kind of place they were sending their son into[s3][s4].

Ten-year-old Huang was the youngest boarder in the school's history. His experience there amounted to an ordeal. His roommate was a seventeen-year-old delinquent covered in tattoos and knife scars, fresh out of prison. Huang taught the illiterate roommate to read and write; in exchange, the roommate taught him to bench-press. Every day he had to clean the toilets of the boys' dormitory—a jarring contrast with the expectations of his mother, a teacher. His older brother, being older, was sent for manual labour on a nearby tobacco farm. As "a short Asian immigrant with long hair and a heavy English accent," he was often bullied and beaten. He faced racial slurs every day—"Chinks," an insult aimed at Chinese people—and as he crossed the footbridge, other boys would shake the ropes, trying to shake him off. He later recalled: "There was no counsellor to talk to back then. You just had to toughen up and keep going." The experience forged his exceptionally high "tolerance for pain"—later a core concept in his management philosophy[s3][s4][s23].

Yet amid the adversity there was redemption. Huang joined the swimming team and learned to play table tennis; sport became his way of surviving hardship and building confidence. At fourteen he appeared in Sports Illustrated—a Taiwanese immigrant boy who, in a mountain school in Kentucky, had fought his way into the mainstream American sports press with a ping-pong paddle. He later said his memories of Kentucky were "more vivid than any other memory." In 2019, by then hugely successful, he held no grudge and donated US$2 million to the school to build a girls' dormitory named Jen-Hsun Huang Hall—not revenge but gratitude. The resilience learned in those Kentucky mountains ran deeper than any business-school curriculum[s1][s3].

About two years after arriving in Kentucky, around 1975, his parents finally emigrated to Beaverton, Oregon, and the family was reunited. Huang entered Aloha High School, where his academic performance soared; he skipped two grades and graduated at sixteen. He became a nationally ranked table-tennis player and an active member of the mathematics, computer and science clubs. In 1977 the school bought an Apple II, among the most advanced personal computers of the day. Huang wrote his own version of the Snake game in BASIC and played the text game Super Star Trek. It was his first close encounter with the world of computers, though he did not yet know it would be the meeting that changed his life[s1].

From the age of fifteen (roughly 1978 to 1983), Huang worked the graveyard shift at a local Denny's restaurant—dishwasher, busboy, waiter, whatever needed doing. He later said he had "washed a lot, a lot of toilets." From Kentucky toilets to the Denny's kitchen, his life seemed to keep intersecting with toilets—but it was precisely that bottom-rung labour that shaped the pragmatism he later brought to managing a company of tens of thousands. Denny's was "quieter than home, and the coffee was cheap," and you could sit for a long time. He developed a special attachment to the chain; years later, he would sketch NVIDIA's blueprint in a Denny's booth[s5][s6].

In 1979, at sixteen, Huang was admitted to Oregon State University to study electrical engineering. His reason for choosing the school was pragmatic—"in-state tuition was cheap." In a university laboratory he met Lori Mills; of the 250 students in the electrical-engineering department at the time, only three were women, and Lori was one of them. The two were lab partners, and shared courses and interests slowly drew them together. He promised Lori that he would "become the CEO of a company by the time I was thirty"—a pledge that seemed fanciful to others but was later kept, word for word[s15][s16].

### 2. 从AMD到LSI Logic：芯片行业的学徒时代 / From AMD to LSI Logic: An Apprenticeship in the Chip Industry

1984年，21岁的黄仁勋从俄勒冈州立大学毕业，获得电子工程学士学位（BSEE），随即加入AMD（超威半导体）担任芯片设计师。在AMD，他开始学普通话——目的是和公司里的中国光罩工人沟通。他通过与同事的日常对话，用语音方式学习中文，这段经历为他日后与中国市场打交道奠定了独特的语言基础。在AMD的时间虽然不长，但让他第一次深入了解了芯片设计的工业流程——从设计到光罩、从流片到量产，每一个环节都需要精密的协调和无数次的迭代。这种对制造环节的深刻理解，后来成为英伟达区别于许多纯设计公司的竞争优势之一[s1][s5]。

1985年，黄仁勋转职至LSI Logic公司——这是一家在 ASIC（专用集成电路）领域颇有建树的半导体公司。在LSI Logic的岁月里，他不仅在技术上持续成长，更重要的是经历了一次关键的职业转型：被调到了销售部门工作过一段时间。他后来认为这是"最佳职业选择之一"——销售经历让他学会了从客户角度思考问题，理解市场需求而非仅仅追求技术完美。一个工程师学会了卖东西，一个技术人学会了理解商业——这种跨界能力成为他日后执掌英伟达的核心竞争力之一。很多纯技术背景的CEO最终失败，不是因为他们不够聪明，而是因为他们无法理解客户的真实需求[s5][s15]。

更重要的是，在LSI Logic，黄仁勋结识了两位日后共同创办英伟达的同事——Chris Malachowsky和Curtis Priem。三人之间建立了深厚的信任和技术默契。Priem是一位技术远见者，对图形处理技术有着超前的理解，1994年就预言"图形处理器的晶体管数量会超过CPU，因为并行计算是未来"——当时被业内视为天方夜谭。Malachowsky则是运营方面的能手，对工程团队的日常管理和生产制造有着丰富经验，是那种能把远景落地为产品的人。黄仁勋则是三人中唯一做过销售的人，懂得如何讲故事、拉投资、建关系。三个人的能力完美互补——技术远见+工程运营+商业战略[s7][s8]。

工作之余，黄仁勋利用业余时间攻读斯坦福大学硕士学位。1990年，27岁的他获得斯坦福大学电子工程硕士学位（MSEE），期间系统接触了图形渲染技术——这在当时还是一个相对小众的领域，研究的人不多，商业前景更是不被看好。但黄仁勋敏锐地意识到，随着PC游戏产业的兴起和多媒体应用的普及，图形处理将成为计算领域中越来越重要的环节。这个判断后来被证明是完全正确的[s1][s7]。

在LSI Logic的八年（1985-1993），黄仁勋从一个初出茅庐的工程师成长为有管理经验的资深从业者。他对芯片行业的商业模式、客户关系和竞争格局有了全面而深入的理解。更重要的是，他与Priem和Malachowsky之间关于创业可能性的讨论，从模糊的想法逐渐变为具体的计划。三人经常在一起讨论3D图形加速的市场机会——当时PC游戏正在兴起，而CPU处理图形的能力远远跟不上需求。他们看到了一个巨大的空白：没有人专门做图形加速芯片。这不是一个已经存在的市场——这正是黄仁勋后来说的"零亿美元的市场"[s7][s8][s18]。

1992-1993年，三人的讨论进入实质阶段。他们经常在加州圣何塞东部的一家Denny's餐厅碰头，讨论创业计划。选择Denny's的原因很务实：黄仁勋年轻时在Denny's打过工，对这里有亲切感；Denny's"比家里安静，咖啡便宜"，可以坐很久讨论商业计划；而且这是三人都负担得起的创业场所——他们没有钱去租办公室或在高级餐厅谈事。三人中，黄仁勋最年轻（30岁），另外两位都比他年长。当被问到为什么他当CEO时，黄仁勋后来笑道："因为他们不想要这份工作。"在剑桥大学演讲中他又补充道："回想起来，我本可以更聪明一些——当CEO是一生的牺牲。"[s6][s8][s21]

**English:** In 1984, at twenty-one, Huang graduated from Oregon State University with a bachelor's degree in electrical engineering (BSEE) and immediately joined AMD, Advanced Micro Devices, as a chip designer. At AMD he began learning Mandarin—so that he could communicate with the company's Chinese mask workers. He picked up the language phonetically through daily conversation with colleagues, an experience that gave him a distinctive linguistic foundation for dealing with the China market in years to come. His time at AMD was short, but it gave him his first deep understanding of the industrial workflow of chip design—from design to masks, from tape-out to mass production, every stage requiring precise co-ordination and countless iterations. That intimate grasp of manufacturing would later become one of the competitive advantages separating NVIDIA from many pure-play design houses[s1][s5].

In 1985 Huang moved to LSI Logic, a semiconductor company with a strong record in ASICs, application-specific integrated circuits. At LSI Logic he continued to grow technically, but more importantly he underwent a pivotal career shift: he was assigned for a time to the sales department. He later called it "one of the best career choices I ever made." Sales taught him to think from the customer's point of view and to understand market demand rather than merely pursuing technical perfection. An engineer who had learned to sell, a technologist who had learned to understand business—that cross-disciplinary ability became one of his core strengths in running NVIDIA. Many CEOs from purely technical backgrounds fail not because they are insufficiently intelligent but because they cannot grasp what customers truly need[s5][s15].

More importantly, at LSI Logic Huang met the two colleagues with whom he would later found NVIDIA: Chris Malachowsky and Curtis Priem. The three built deep trust and technical rapport. Priem was a technical visionary with an ahead-of-the-curve understanding of graphics processing; as early as 1994 he predicted that graphics processors would contain more transistors than CPUs, because parallel computing was the future—a view the industry then dismissed as fantasy. Malachowsky was the operations man, richly experienced in the day-to-day management of engineering teams and in manufacturing, the kind of person who could turn a vision into a shippable product. Huang was the only one of the three who had worked in sales; he knew how to tell a story, raise money and build relationships. Their capabilities complemented one another perfectly: technical vision, engineering operations and business strategy[s7][s8].

Outside work, Huang used his spare time to pursue a master's degree at Stanford University. In 1990, at twenty-seven, he earned a master's in electrical engineering (MSEE) from Stanford, where he was systematically exposed to graphics-rendering technology—then a relatively niche field, with few researchers and even fewer believers in its commercial prospects. But Huang sensed acutely that, with the rise of the PC-gaming industry and the spread of multimedia applications, graphics processing would become an ever more important part of computing. The judgment proved entirely correct[s1][s7].

During his eight years at LSI Logic (1985-1993), Huang grew from a fledgling engineer into a seasoned professional with management experience. He gained a comprehensive, deep understanding of the chip industry's business model, customer relationships and competitive landscape. Above all, his discussions with Priem and Malachowsky about the possibility of starting a company evolved from a vague idea into a concrete plan. The three often talked about the market opportunity in 3D graphics acceleration: PC gaming was taking off, and CPUs were nowhere near fast enough at processing graphics. They saw a huge gap: no one was building dedicated graphics-acceleration chips. This was not an existing market at all—it was exactly what Huang later called a "zero-billion-dollar market"[s7][s8][s18].

In 1992-1993 the discussions turned substantive. They met often at a Denny's restaurant in east San Jose to plan the venture. The choice of Denny's was pragmatic: Huang had worked there as a young man and felt an attachment; it was "quieter than home, and the coffee was cheap," so they could sit for hours on a business plan; and it was a venue the three could afford—they had no money to rent an office or meet over fine dining. Of the three, Huang was the youngest at thirty; the other two were older. Asked later why he became CEO, Huang laughed: "Because they didn't want the job." In his Cambridge speech he added: "In hindsight, I could have been smarter—to be a CEO is a lifetime of sacrifice."[s6][s8][s21]

### 3. Denny's创业：4万美元与三个工程师的赌注 / Founding at Denny's: US$40,000 and the Bet of Three Engineers

1993年4月5日，黄仁勋与Chris Malachowsky、Curtis Priem共同创立NVIDIA（英伟达）。创业地点在加州圣何塞东部的一家Denny's餐厅——这家美式连锁餐厅后来几乎成为英伟达创业神话的标志性符号，就像乔布斯的车库、比尔·盖茨的哈佛宿舍一样。黄仁勋时年30岁，出任总裁兼CEO，兑现了多年前对妻子Lori的承诺。需要澄清的是，公司成立日（4月5日）并非他的生日（2月17日），但"在30岁那年创立公司"本身就已足够传奇[s7][s8]。

创始团队的分工明确而互补，堪称经典创业组合。黄仁勋任总裁兼CEO，负责战略、融资、客户关系和整体管理——他是三人中唯一做过销售的人，这使他成为CEO的天然人选。在LSI Logic的销售经历让他知道如何向投资人讲一个让人信服的故事，如何理解客户的需求而不是自说自话。Chris Malachowsky主管工程运营，是NVIDIA的运营支柱，负责工程团队的日常运营、生产制造和质量控制——他是那种确保产品能从设计图纸变成实物的关键人物。Curtis Priem任首席技术官（CTO），是公司的技术远见者——他不仅看到了3D图形的未来，更看到了并行计算将彻底改变计算机架构的深层趋势[s8]。

创业资金极为紧张，与后来英伟达的万亿帝国形成了戏剧性的对比。约4万美元的银行贷款作为启动资金，三人共同筹措。另一个广为流传的细节是：公司注册时律师要了黄仁勋口袋里仅有的200美元现金，他又向两位合伙人各要了200美元——NVIDIA的"名义注册资本"只有600美元。这个故事听上去有些心酸，但实际运营靠的是更正规的后续融资：1993年从红杉资本（Sequoia Capital）和Sutter Hill Ventures获得200万美元种子轮融资，投后估值600万美元。后续多轮风险投资总计约2,000万美元，包含世嘉（Sega）500万美元战略投资。红杉资本合伙人Mark Stevens后来总结了黄仁勋的一句名言："我们投资于零亿美元的市场。"——这句话后来成为英伟达战略哲学的核心表达[s8][s9][s18]。

公司名字的由来颇具戏剧性。他们考虑过很多候选名——PixelPushers（太直白，像个作坊的名字）、Rendition（已被另一家图形公司使用）等等。最终选择"NVIDIA"——源自拉丁语"invidia"，意为"嫉妒"（罗马神话中的嫉妒女神Invidia）。发音为en-VID-ee-ah。绿色眼睛Logo，螺旋图案代表无限计算，暗含"要让同行仰望、嫉妒"的勃勃野心。早期文件拼写为"nVidia"（小写n），1997年后统一为全大写的"NVIDIA"——这个小小的字体变化，暗示着公司从初创到成熟的心理转变[s1]。

创始愿景可以归纳为三根支柱，分别来自三位创始人。技术愿景由黄仁勋主导——"我们要创造一个全新的计算类别，让视觉计算成为继CPU之后最重要的处理器"。市场愿景来自Malachowsky——将3D图形带入游戏和多媒体市场，这是一个正在快速膨胀的市场。远见来自Priem——1994年就预言GPU的晶体管数量将超过CPU，因为并行计算是未来。这个在当时被大多数人视为疯话的预测，在二十年后成为现实[s7][s8]。

创业初期的混乱是可以想象的。三人"完全不知道怎么开公司"。没有正式的商业计划书，没有西装革履的路演PPT，就在Denny's的卡座里用铅笔在餐巾纸上画下了公司的蓝图。黄仁勋后来说："如果我当时就知道今天所知道的一切，我不会再创办这家公司。太可怕了，太痛苦了，牺牲太大了。"这句话道出了创业的真实代价，也暗含着一种 retrospective 的感慨——正是因为不知道前方有多难，才敢迈出那一步。无知，有时候是一种祝福[s21]。

**English:** On April 5, 1993, Jensen Huang co-founded NVIDIA with Chris Malachowsky and Curtis Priem. The company was born in a Denny's restaurant in east San Jose, California—a diner that would become nearly as emblematic of NVIDIA's founding myth as Steve Jobs's garage or Bill Gates's Harvard dormitory. Huang was thirty years old and became president and CEO, keeping the promise he had made to his wife Lori years before. For the record, the company's founding date (April 5) is not his birthday (February 17), but "starting a company in his thirtieth year" is legendary enough on its own[s7][s8].

The founding team's division of labour was clear and complementary—a classic startup configuration. Huang served as president and CEO, responsible for strategy, fundraising, customer relationships and overall management; the only one of the three with sales experience, he was the natural choice for the job. His sales work at LSI Logic had taught him how to tell investors a convincing story and how to understand customers rather than talk to himself. Chris Malachowsky led engineering operations, NVIDIA's operational backbone, overseeing the engineering teams' day-to-day work, manufacturing and quality control—the crucial figure who ensured products made it from blueprints into physical reality. Curtis Priem became chief technology officer (CTO), the company's technical visionary: he saw not only the future of 3D graphics but the deeper trend that parallel computing would transform computer architecture altogether[s8].

Startup funding was desperately tight, a dramatic contrast with NVIDIA's later trillion-dollar empire. The three scraped together about US$40,000 in bank loans as seed capital. A widely circulated detail: when the company was incorporated, the lawyer took the only US$200 in cash Huang had in his pocket, and Huang then asked each of his two partners for another US$200—NVIDIA's notional registered capital was just US$600. The story sounds poignant, but the actual operation rested on more conventional later financing: in 1993 the company raised a US$2 million seed round from Sequoia Capital and Sutter Hill Ventures, at a post-money valuation of US$6 million. Subsequent venture rounds totalled roughly US$20 million, including a US$5 million strategic investment from Sega. Sequoia partner Mark Stevens later distilled one of Huang's signature lines: "We invest in zero-billion-dollar markets." It became the central expression of NVIDIA's strategic philosophy[s8][s9][s18].

The company name had a dramatic genesis. They considered many candidates—PixelPushers (too blunt, sounded like a workshop), Rendition (already taken by another graphics company) and more. In the end they chose "NVIDIA," from the Latin "invidia," meaning envy (Invidia, the Roman goddess of envy). It is pronounced en-VID-ee-ah. The green-eyed logo, with its spiral suggesting infinite computation, carried an audacious ambition: to make rivals look up in envy. Early documents spelled it "nVidia" with a lowercase n; after 1997 it was standardised as the all-caps "NVIDIA"—a small typographic shift that hints at the company's psychological transition from startup to maturity[s1].

The founding vision can be summed up in three pillars, one from each founder. The technical vision was led by Huang: "We were going to create a brand-new computing category, making visual computing the most important processor after the CPU." The market vision came from Malachowsky: bring 3D graphics to gaming and multimedia, a market swelling fast. The far-sighted prediction came from Priem, who in 1994 foresaw that GPUs would contain more transistors than CPUs because parallel computing was the future. A forecast most people then dismissed as crazy became reality two decades later[s7][s8].

The chaos of the early days was predictable. The three had "absolutely no idea how to run a company." There was no formal business plan, no suit-and-tie roadshow deck; in a Denny's booth they sketched the company's blueprint in pencil on napkins. Huang later said: "If I had known then everything I know today, I would never have started the company. It was terrifying, it was painful, the sacrifices were enormous." The line captures the true price of entrepreneurship, with a retrospective twist: it was precisely because they did not know how hard the road ahead would be that they dared to take the first step. Ignorance, sometimes, is a blessing[s21].

### 4. 生死一线：NV1失败、世嘉赌局与RIVA逆转 / Between Life and Death: The NV1 Failure, the Sega Gamble and the RIVA Turnaround

创业的头两年，NVIDIA几乎就经历了灭顶之灾——这对于一家后来市值5万亿美元的公司来说，几乎是不可想象的。

1995年推出的首款芯片NV1采用了四边形渲染技术路线（quad rendering），这在当时的技术选择中并非没有道理——四边形在某些图形应用中确实有优势。但NV1犯了一个致命错误：不兼容微软Direct3D标准（行业标准正转向三角形渲染）。在PC图形领域，微软的标准就是游戏规则，不兼容Direct3D就等于被市场抛弃。几乎没有游戏开发商愿意为NV1写驱动程序，产品几乎卖不出去[s5][s9]。

NV1的失败将公司推入绝境。1996年，NVIDIA银行账户里不足100万美元。团队从约100人裁至约30-35人，裁员幅度约70%——这意味着黄仁勋不得不亲手送走他亲手招募的同事和朋友。黄仁勋抵押了自己的房产为员工发工资。这是他人生中最接近彻底失败的一次——后来他反复说的那句"我们距离倒闭永远只有30天"，其源头就在这里。这不是一句修辞，而是来自切肤之痛的真实体验[s9][s19]。

就在濒临破产之际，一丝曙光出现了。NVIDIA与日本游戏巨头世嘉（Sega）签订了为Dreamcast游戏机设计图形芯片的合同——这是一笔足以改变公司命运的大单。然而，黄仁勋很快发现了一个致命问题：他们正在开发的架构是技术死胡同——这条技术路线无法扩展，如果硬着头皮做完，产品性能将无法满足市场需求，不仅世嘉会不满意，NVIDIA也会名誉扫地，彻底失去翻盘机会[s9][s10]。

接下来发生的事成为商业史上最经典的"弱方谈判"案例之一。黄仁勋飞赴日本面见世嘉社长，主动告诉对方——NVIDIA要停止开发这款芯片。这等于承认自己做不出来，等于亲手毁掉了一份救命的合同。然后，他提出了一个更惊人的要求：请求世嘉仍然支付全额合同款——即使世嘉永远不会收到这款产品。这简直是在要求对方做一件完全不合理的事[s10]。

根据《The Closer》的深度报道，黄仁勋说服世嘉CEO的关键不是论据，而是信念。他对技术方向的绝对确信——他确信三角形渲染才是未来，确信NVIDIA能做出更好的东西——本身就具有说服力。他不是在从实力地位谈判（他根本没有实力），而是在从信念地位谈判。世嘉CEO被他打动，同意了这笔交易。这笔钱（有说500万美元战略投资，有说约700万美元合约金）拯救了NVIDIA。黄仁勋后来说："如果他说不，我们就会倒闭。"世嘉最终从这笔投资中间接收回了成本并获利约1,000万美元——虽然Dreamcast最终使用了NEC制造的芯片，但世嘉在这笔交易中并没有亏钱[s9][s10]。

从这次经历中，黄仁勋提炼出两条终生受用的道理：第一，当你没有筹码时，信念就是你唯一的筹码——不是虚张声势，而是真正的、基于深刻技术理解的信念；第二，承认错误并及时止损，比硬撑到底更需要勇气——如果他没有勇气告诉世嘉"我们做错了"，NVIDIA就会在一条死胡同里走到黑[s10]。

用世嘉的钱，NVIDIA全力开发了RIVA 128（NV3）。这一次，他们做对了——128位3D图形加速，完全兼容Direct3D标准。1997年4月推出后，仅四个月就卖出百万块，公司首度实现盈利。财务数据的变化令人惊叹：1996年营收仅350万美元（净亏损950万美元），1997年跃升至5,750万美元（净利润130万美元），员工从35人增至85人。从濒死到盈利，只用了不到一年时间[s5][s9]。

这次死里逃生的经历，深深影响了黄仁勋此后的每一次重大决策。他学会了在所有人都说"继续做"的时候，问一个更本质的问题："这条路走得通吗？"如果答案是否定的，无论已经投入了多少，都要有勇气止损。后来在CUDA的十年豪赌中，在放弃400亿美元收购Arm交易中，都可以看到这种"止损勇气"的影子[s10][s19]。

**English:** In its first two years NVIDIA came close to total catastrophe—almost unimaginable for a company that would one day be worth US$5 trillion.

The first chip, the NV1, launched in 1995, used a quadrilateral-rendering (quad rendering) approach, a choice that was not without technical rationale: quads did offer advantages in certain graphics applications. But the NV1 made a fatal error: it was incompatible with Microsoft's Direct3D standard, as the industry was shifting to triangle rendering. In PC graphics, Microsoft's standard was the rules of the game; failing to support Direct3D amounted to being abandoned by the market. Hardly any game developers were willing to write drivers for the NV1, and the product barely sold[s5][s9].

The NV1's failure pushed the company to the brink. In 1996 NVIDIA had less than US$1 million in the bank. The workforce was cut from roughly 100 people to about 30 to 35—a reduction of around 70 percent, meaning Huang had to personally send away colleagues and friends he had personally recruited. He mortgaged his own house to make payroll. It was the closest he ever came to outright failure, and it is the source of the line he has repeated ever since: "We are always thirty days away from going out of business." It is not rhetoric; it is a truth learned in the flesh[s9][s19].

Just as bankruptcy loomed, a sliver of light appeared. NVIDIA signed a contract with the Japanese gaming giant Sega to design the graphics chip for the Dreamcast console—a deal large enough to change the company's fate. Before long, however, Huang spotted a fatal problem: the architecture they were developing was a technical dead end. The approach could not scale; if they forced it through to completion, the product's performance would fail to meet market demand, Sega would be dissatisfied, and NVIDIA's reputation would be ruined, destroying any chance of recovery[s9][s10].

What followed became one of the classic "weak-hand negotiations" in business history. Huang flew to Japan to meet Sega's president and told him, unprompted, that NVIDIA was halting development of the chip. This was an admission that they could not build it—a move that amounted to tearing up a life-saving contract with his own hands. Then he made an even more astonishing request: he asked Sega to pay the full contract price anyway, even though Sega would never receive the product. It was, on its face, a demand that the other side do something utterly unreasonable[s10].

According to in-depth reporting in The Closer, what persuaded Sega's CEO was not argument but conviction. Huang's absolute certainty about the technical direction—his certainty that triangle rendering was the future and that NVIDIA could build something better—was itself persuasive. He was not negotiating from a position of strength (he had none); he was negotiating from a position of conviction. Sega's CEO was moved and agreed to the deal. The money—variously reported as a US$5 million strategic investment or around US$7 million in contract payments—saved NVIDIA. Huang later said: "If he had said no, we would have gone out of business." Sega ultimately recovered its cost indirectly and made about US$10 million on the investment; although the Dreamcast eventually used a chip made by NEC, Sega did not lose money on the transaction[s9][s10].

From the experience Huang distilled two lessons that served him for life. First, when you have no chips to play, conviction is your only chip—not bluster, but genuine conviction grounded in deep technical understanding. Second, admitting a mistake and cutting losses in time takes more courage than stubbornly holding on; had he lacked the nerve to tell Sega "we got it wrong," NVIDIA would have walked a dead-end road to the end[s10].

With Sega's money, NVIDIA poured everything into the RIVA 128 (NV3). This time they got it right: 128-bit 3D graphics acceleration, fully compatible with the Direct3D standard. Launched in April 1997, it sold a million units in just four months and the company turned a profit for the first time. The financial transformation was striking: revenue of only US$3.5 million in 1996 (with a net loss of US$9.5 million) jumped to US$57.5 million in 1997 (net income of US$1.3 million), while headcount rose from 35 to 85. From near-death to profitability took less than a year[s5][s9].

That escape from extinction shaped every major decision Huang made afterward. He learned to ask, when everyone else was saying "keep going," a more fundamental question: "Does this road actually work?" If the answer was no, no matter how much had already been invested, he had the courage to stop. The same "courage to cut losses" can be seen later in the decade-long CUDA gamble and in abandoning the US$40 billion Arm acquisition[s10][s19].

### 5. 发明GPU与上市：从GeForce到纳斯达克 / Inventing the GPU and Going Public: From GeForce to NASDAQ

1999年是英伟达历史上里程碑式的一年，也是整个计算机图形学历史上的分水岭。这一年，公司推出了GeForce 256，首次提出"GPU"（图形处理器，Graphics Processing Unit）这一全新概念——这不是一个简单的产品命名，而是定义了一个全新的计算品类。在此之前，人们只把图形加速卡当作CPU的附属品，一个"帮CPU画图"的配角。GeForce 256改变了这一切。它集成了变换与光照（T&L）硬件加速，这是一项革命性创新——在此之前，3D图形的变换和光照计算需要CPU来完成，而GPU的出现将这一重担从CPU转移到了专用处理器上，彻底改变了计算机图形学的架构范式[s7][s11]。

"GPU"这个概念的提出，是黄仁勋"品类创造"思维的经典体现。他不是在做一块"更快的显卡"，而是在创造一个全新的东西，给它一个全新的名字，定义一套全新的标准。这种思维方式后来贯穿了英伟达的整个发展史——从CUDA到PhysX，从光线追踪到Tensor Core，每一次重大创新都伴随着一个新概念的提出[s11]。

同年1月22日，英伟达在纳斯达克成功上市，股票代码NVDA。发行价12美元/股，发行350万股，募集资金约4,200万美元，IPO估值约6亿美元。上市时公司拥有约250名员工，1998年全年营收1.58亿美元，1999年上半年已达1.12亿美元。从4万美元的启动资金到6亿美元的上市市值，英伟达只用了不到六年时间——这个速度在90年代末的硅谷也是令人瞩目的[s9][s11]。

2000年，英伟达接连拿下两个关键成就，奠定了在图形芯片领域不可撼动的领导地位。第一，微软选择NVIDIA为初代Xbox游戏主机供应GPU——这是一个战略性的大单，意味着英伟达的图形技术将成为全球数百万游戏玩家的标准体验。第二，以约7,000万美元收购了曾经的竞争对手3dfx的核心资产。3dfx是GPU领域的先驱之一，其Voodoo系列显卡曾是PC游戏的标配，是无数玩家心中"高性能"的代名词，但公司经营不善走向衰落。这次收购一统了显卡市场的格局——从此在独立GPU领域，NVIDIA的主要竞争对手只剩下ATI（后来被AMD收购）[s11]。

上市后的英伟达进入了快速迭代期。公司确立了"双轮驱动"的架构节奏：大约每两年推出一代全新架构，每一代在性能和功能上实现显著跃升。这种节奏延续至今——从GeForce 256到GeForce 3，从GeForce FX到GeForce 8系列，从Tesla到Fermi，从Kepler到Maxwell，从Pascal到Volta，从Turing到Ampere，从Hopper到Blackwell，再到Vera Rubin——每一代架构的发布都是全球科技界的重大事件。这种持续的技术迭代能力，是英伟达能够长期保持领先的关键[s7][s11]。

更重要的是，GeForce系列产品的成功不仅为公司带来了丰厚的利润，还创造了一个庞大的"安装基数"——数以百万计的GeForce显卡分布在世界各地的工作站、游戏PC和数据中心里。黄仁勋后来意识到，这个安装基数可以成为远超游戏领域的战略资产——它是CUDA生态系统的物理基础，是AI革命的硬件底座。正如他所说的："安装基数定义一种架构。"当全世界有数百万块显卡都支持同一种计算架构时，开发者就会围绕这种架构写软件，用户就会购买支持这种架构的硬件——一个强大的飞轮效应就此形成[s10][s12]。

**English:** 1999 was a landmark year in NVIDIA's history and a watershed in the entire history of computer graphics. That year the company launched the GeForce 256 and introduced the brand-new concept of the "GPU," the graphics processing unit—not a simple product name but the definition of an entirely new computing category. Before then, graphics accelerators had been treated as accessories to the CPU, a supporting player that "helped the CPU draw pictures." The GeForce 256 changed all that. It integrated hardware-accelerated transform and lighting (T&L), a revolutionary innovation: previously, the transform and lighting calculations for 3D graphics had to be performed by the CPU, and the arrival of the GPU shifted that burden from the CPU onto a dedicated processor, fundamentally changing the architectural paradigm of computer graphics[s7][s11].

The coinage of "GPU" is a classic expression of Huang's "category creation" mindset. He was not building a "faster graphics card"; he was creating something entirely new, giving it a new name and defining a new set of standards. That mode of thinking would run through NVIDIA's entire history—from CUDA to PhysX, from ray tracing to Tensor Cores, every major innovation accompanied by the introduction of a new concept[s11].

On January 22 of the same year, NVIDIA listed successfully on NASDAQ under the ticker NVDA. The offering was priced at US$12 a share, with 3.5 million shares issued, raising about US$42 million at an IPO valuation of roughly US$600 million. At listing the company had about 250 employees; full-year 1998 revenue was US$158 million, and the first half of 1999 alone had reached US$112 million. From US$40,000 in startup money to a US$600 million market capitalisation at IPO took NVIDIA less than six years—a pace that commanded attention even in late-1990s Silicon Valley[s9][s11].

In 2000 NVIDIA secured two pivotal achievements that cemented unshakeable leadership in graphics chips. First, Microsoft chose NVIDIA to supply the GPU for the original Xbox console—a strategic blockbuster, meaning NVIDIA's graphics technology would become the standard experience for millions of gamers worldwide. Second, for about US$70 million it acquired the core assets of its onetime rival 3dfx. 3dfx had been one of the pioneers of the GPU field; its Voodoo line of cards was once standard equipment in PC gaming and the byword for "high performance" for countless players, before mismanagement drove the company into decline. The acquisition unified the graphics-card landscape: henceforth, in discrete GPUs, NVIDIA's only major rival was ATI (later acquired by AMD)[s11].

After going public NVIDIA entered a period of rapid iteration. The company established a "two-wheel drive" architectural cadence: roughly every two years it introduced a wholly new generation of architecture, each delivering a substantial leap in performance and features. That rhythm continues to this day—from GeForce 256 to GeForce 3, from GeForce FX to the GeForce 8 series, from Tesla to Fermi, from Kepler to Maxwell, from Pascal to Volta, from Turing to Ampere, from Hopper to Blackwell and on to Vera Rubin—each architecture launch a major event in global technology. This sustained capacity for iteration is the key to NVIDIA's long-running lead[s7][s11].

More importantly, the success of the GeForce line not only brought the company handsome profits but created a vast installed base: millions of GeForce cards distributed across workstations, gaming PCs and data centres around the world. Huang later realised that this installed base could become a strategic asset reaching far beyond gaming—it was the physical foundation of the CUDA ecosystem and the hardware bedrock of the AI revolution. As he put it: "The installed base defines an architecture." When millions of cards worldwide support the same computing architecture, developers write software around that architecture and users buy hardware that supports it—a powerful flywheel set in motion[s10][s12].

### 6. CUDA十年孤注一掷：从"疯子"到AI基石 / CUDA: A Decade of Betting the Company, From "Madman" to Cornerstone of AI

CUDA的故事不是从黄仁勋的顶层设计开始的，而是从一个博士生的"黑客"行为开始的——这本身就充满了隐喻意味：最伟大的创新，往往来自体制之外的"非预期用途"。

2004年，斯坦福大学计算机科学博士生Ian Buck在用两块NVIDIA显卡做一件奇怪的事——他开发了名为Brook的编程语言，让开发者可以绕过图形管线，直接对GPU进行通用计算。在传统的理解中，GPU就是用来画图的——你不能指望一块显卡去做科学计算，就像你不能指望一把菜刀去砍树一样。但Ian Buck做到了，至少做到了原型验证。NVIDIA注意到了他的工作，2004年将他招入麾下。在NVIDIA，Buck与GPU计算架构总监John Nickolls合作，将研究原型转化为产品——这就是CUDA的由来[s12]。

黄仁勋敏锐地看到了这个现象背后的深层含义。他注意到，大学里的研究者正在"黑"他的显卡做科学计算——物理模拟、分子建模、气候模拟、金融工程。这些研究者的行为完全是对GPU的"非预期使用"，他们是在与显卡"搏斗"中完成了科学计算。如果人们愿意费这么大劲、克服这么多困难去用图形芯片做计算，也许计算的未来本身就是并行的。大多数公司会忽略这种"非预期用途"——它太小了，太边缘了，不符合任何市场调研报告的结论。但黄仁勋看到了一个模式[s10][s12]。

2006年，CUDA正式发布（2007年正式推出）。CUDA（Compute Unified Device Architecture，统一计算设备架构）允许开发者用C语言等通用编程语言直接调用GPU算力进行通用计算——GPU从此可以用于游戏之外的任何计算任务。这是一个划时代的创新：它打破了GPU只能做图形的限制，将GPU变成了通用的并行计算加速器[s12]。

但黄仁勋做了一个更为大胆的决定——不仅是在专业工作站GPU上支持CUDA，而是在每一块GeForce游戏显卡上都植入CUDA功能，包括最便宜的消费级显卡。他的逻辑是："安装基数定义架构。"（The installed base defines an architecture.）如果CUDA要成为一个新的计算平台，它必须被尽可能多的人掌握。只有当全世界有几百万、几千万块支持CUDA的显卡时，开发者才会愿意为它写软件——因为他们知道，写完的代码可以在无数台机器上运行[s10][s12]。

这个决定的代价是毁灭性的。每块GeForce显卡的生产成本增加约50%——那些最便宜的、利润微薄的游戏卡，现在要承载原本不属于它们的计算功能。整体毛利率暴跌至约35%，每颗芯片的成本增加约1.5美元。华尔街疯狂抛售——英伟达市值从约80-120亿美元暴跌至约15-20亿美元，跌幅约80%。分析师质疑黄仁勋"是不是疯了"——一个游戏显卡公司为什么要投资科学计算软件？这是"不务正业"。消费市场只关心游戏画面是否更流畅，根本不在乎显卡能否做分子动力学模拟。公司内部也有反对声音，游戏部门不理解为什么要为一个"不存在的市场"增加成本[s10][s12][s24]。

黄仁勋后来称CUDA是"第一个最接近存在性威胁的战略决策"（the first strategic decision that came closest to an existential threat）。他回忆发布时的场景："当我推出CUDA时，观众完全沉默。没有人想要它。没有人要求它。没有人理解它。"这种沉默比反对更可怕——反对意味着至少有人在乎，而沉默意味着你是在对着虚空说话[s10]。

这就是"沉默的十年"的开始。在这十年里，黄仁勋做了大量"播种"工作：亲自去大学教课，编写CUDA教材，向研究人员赠送开发卡，每年投入数亿美元维护和推广CUDA。黄仁勋本人在2024年接受CBS《60分钟》采访时给出过一个口径：“我们年复一年地投资CUDA——十亿美元甚至更多——在其他人还看不出我们为什么这么做的时候。”他后来在多个场合复盘，CUDA前后约15年的持续投入累计约100亿美元——这笔在漫长时间里看不到回报的钱，构成了英伟达最宽的软件护城河，建立CUDA库生态——cuBLAS（基础线性代数）、cuFFT（快速傅里叶变换）、cuDNN（深度神经网络）等等。这些库将GPU的计算能力转化为科学家和工程师可以直接使用的工具。CUDA开发者从2007年的几百人缓慢增长到2010年的数千人。大多数公司会在第三、四年就砍掉这种"看不到回报"的项目，但黄仁勋坚持了整整十年[s12]。

他的坚持基于三条底层逻辑。第一，第一性原理——并行计算是计算的未来，这是物理定律决定的。CPU的串行速度已经接近物理极限（频率墙），而数据量在爆炸式增长，只有并行计算才能应对这种增长。第二，安装基数飞轮——越多显卡支持CUDA，越多开发者写CUDA代码，越多CUDA应用带动更多显卡销售，更多显卡销售带来更多CUDA安装基数——一个自我强化的循环。第三，"零亿美元市场"哲学——投资于一个已经存在的市场，你在和所有能看到同样机会的人竞争，利润被无限压缩。投资于一个还不存在的市场，你有时间建立基础设施、生态系统和护城河——在任何人意识到发生了什么之前[s10][s12][s18]。

2012年，转折点到来了。多伦多大学的Alex Krizhevsky用两块消费级GTX 580显卡训练的AlexNet模型在ImageNet竞赛中取得断层式突破——超过10个百分点的差距。一夜之间，所有严肃的深度学习实验室都开始购买NVIDIA GPU并编写CUDA代码。十年孤独的等待，终于换来了"市场突然存在"的时刻——不是渐进式的，而是爆发式的[s12][s13]。

到2026年，CUDA生态已包含300多个库、600多个AI模型、3,700个GPU加速应用，累计下载量超5,300万次，开发者超过500万。正如分析人士精辟指出的：竞争对手可以复制芯片（很难但不是不可能），但无法复制十年的生态积累——全世界的AI研究者已经花了十年时间写CUDA代码、构建CUDA库、用CUDA训练学生。每一篇深度学习论文的参考文献里，都藏着CUDA的代码。硬件赢得了赛跑，软件赢得了战争，而战争是复利的[s12]。

**English:** The CUDA story did not begin with top-down design by Jensen Huang. It began with a doctoral student's act of "hacking"—which is itself full of metaphor: the greatest innovations often come from "unintended uses" outside the establishment.

In 2004, Ian Buck, a computer-science PhD student at Stanford, was doing something strange with two NVIDIA graphics cards: he had developed a programming language called Brook that let developers bypass the graphics pipeline and perform general-purpose computing directly on the GPU. In the conventional understanding, GPUs were for drawing pictures—you could no more expect a graphics card to do scientific computing than expect a kitchen knife to fell a tree. But Buck did it, at least to the proof-of-concept stage. NVIDIA noticed his work and hired him in 2004. At NVIDIA, Buck worked with John Nickolls, director of GPU computing architecture, to turn the research prototype into a product—and that was the origin of CUDA[s12].

Huang saw acutely the deeper meaning behind the phenomenon. He noticed that university researchers were "hacking" his graphics cards to do scientific computing—physics simulations, molecular modelling, climate simulation, financial engineering. Their behaviour was an entirely "unintended use" of the GPU; they were accomplishing scientific computing while wrestling with the cards. If people were willing to go to such lengths and overcome so many obstacles to compute on graphics chips, perhaps the future of computing itself was parallel. Most companies would have ignored such "unintended uses"—too small, too marginal, contradicting every market-research report. But Huang saw a pattern[s10][s12].

In 2006 CUDA was announced (formally launched in 2007). CUDA, the Compute Unified Device Architecture, allowed developers to use general-purpose languages such as C to call on GPU horsepower directly for general-purpose computing—henceforth the GPU could be used for any computational task beyond gaming. It was an epoch-making innovation: it broke the restriction that GPUs could only do graphics and turned the GPU into a general-purpose parallel-computing accelerator[s12].

But Huang made an even bolder decision: CUDA would be supported not only on professional workstation GPUs but built into every GeForce gaming card, including the cheapest consumer models. His logic: "The installed base defines an architecture." If CUDA was to become a new computing platform, it had to be mastered by as many people as possible. Only when millions, tens of millions of CUDA-capable cards existed worldwide would developers be willing to write software for it—because they would know the code they wrote could run on countless machines[s10][s12].

The cost of that decision was devastating. Production cost per GeForce card rose by roughly 50 percent—the cheapest, thinnest-margin gaming cards now carried computing functions never meant for them. Overall gross margin plunged to about 35 percent, and the cost per chip rose by roughly US$1.50. Wall Street sold off furiously: NVIDIA's market capitalisation collapsed from about US$8-12 billion to roughly US$1.5-2 billion, a drop of around 80 percent. Analysts questioned whether Huang had "lost his mind"—why would a gaming-graphics company invest in scientific-computing software? It was "neglecting the day job." The consumer market cared only whether games ran more smoothly, not at all whether a card could run molecular-dynamics simulations. There was internal opposition too; the gaming division could not understand why costs should rise for a "market that didn't exist"[s10][s12][s24].

Huang later called CUDA "the first strategic decision that came closest to an existential threat." He recalled the launch: "When I introduced CUDA, the audience was completely silent. Nobody wanted it. Nobody asked for it. Nobody understood it." That silence was more frightening than opposition: opposition meant at least someone cared, while silence meant you were speaking into the void[s10].

So began the "silent decade." During those years Huang did enormous amounts of "sowing": he personally went to universities to teach, wrote CUDA teaching materials, gave away development cards to researchers, and spent hundreds of millions of dollars every year maintaining and promoting CUDA. In a 2024 interview with CBS's 60 Minutes, Huang gave one accounting: "We invested in CUDA year after year—a billion dollars or more—at a time when nobody else could see why we were doing it." He has reviewed the episode on multiple occasions: roughly fifteen years of sustained CUDA investment totalled about US$10 billion—money that saw no return over a long stretch and yet formed NVIDIA's widest software moat. The company built the CUDA library ecosystem—cuBLAS (basic linear algebra), cuFFT (fast Fourier transforms), cuDNN (deep neural networks) and more—libraries that converted the GPU's computing power into tools scientists and engineers could use directly. CUDA developers grew slowly, from a few hundred in 2007 to several thousand by 2010. Most companies would have killed such a "no-return" project in year three or four; Huang held on for a full decade[s12].

His persistence rested on three underlying logics. First, first principles: parallel computing is the future of computing, determined by the laws of physics. The serial speed of CPUs was approaching physical limits (the frequency wall), while data volumes were exploding; only parallel computing could meet that growth. Second, the installed-base flywheel: the more cards support CUDA, the more developers write CUDA code; the more CUDA applications drive card sales; the more card sales expand the CUDA installed base—a self-reinforcing loop. Third, the "zero-billion-dollar market" philosophy: invest in an existing market and you compete against everyone who can see the same opportunity, with profits compressed to nothing; invest in a market that does not yet exist and you have time to build infrastructure, an ecosystem and a moat—before anyone realises what has happened[s10][s12][s18].

In 2012 the turning point arrived. Alex Krizhevsky of the University of Toronto used two consumer-grade GTX 580 cards to train the AlexNet model, which scored a landslide breakthrough in the ImageNet competition—winning by a margin of more than ten percentage points. Overnight, every serious deep-learning lab began buying NVIDIA GPUs and writing CUDA code. A decade of lonely waiting had finally earned the moment when "the market suddenly existed"—not gradually, but explosively[s12][s13].

By 2026 the CUDA ecosystem encompassed more than 300 libraries, over 600 AI models and 3,700 GPU-accelerated applications, with cumulative downloads exceeding 53 million and more than five million developers. As analysts have shrewdly observed: competitors can copy the chips (hard, but not impossible), but they cannot copy a decade of ecosystem accumulation—AI researchers worldwide have spent ten years writing CUDA code, building CUDA libraries and training students with CUDA. The reference list of every deep-learning paper hides CUDA code. Hardware wins the race; software wins the war; and war is compounded[s12].

### 7. AI大爆炸：从AlexNet到ChatGPT到万亿帝国 / The AI Big Bang: From AlexNet to ChatGPT to the Trillion-Dollar Empire

2012年9月，多伦多大学的博士生Alex Krizhevsky在父母家的卧室里，用两块GeForce GTX 580游戏显卡（每块约500美元），花了5-6天训练了一个深度卷积神经网络——AlexNet。共同作者是Ilya Sutskever和导师Geoffrey Hinton（深度学习三巨头之一）。AlexNet在ImageNet竞赛中取得了15.3%的top-5错误率，而第二名是26.2%——超过10个百分点的差距，不是渐进式改进，而是断层式突破。黄仁勋称之为"AI的大爆炸"（the Big Bang of AI）[s12][s13]。

NVIDIA比几乎所有人都更快地理解了这件事的意义。公司已经观察GPU计算在科学和高性能计算领域的扩展六年了——分子动力学、气候模拟、流体力学。但AlexNet不同。它暗示并行计算最激动人心的应用可能不是物理学，而是智能本身。不是模拟宇宙的行为，而是模拟宇宙中最复杂的结构——大脑。NVIDIA迅速调整公司方向：Tesla产品线重新定位为数据中心计算产品，架构演进（Kepler→Maxwell→Pascal→Volta→Turing→Ampere→Hopper→Blackwell→Rubin）越来越多地针对深度学习工作负载优化[s12]。

2016年，英伟达推出了两项具有里程碑意义的产品和举措。第一是Pascal架构（P100），首款专为深度学习设计的GPU，同时NVLink高速互联技术问世——这意味着GPU之间的通信速度将不再是训练的瓶颈。第二是DGX-1——"箱子里的超级计算机"，搭载8颗Tesla P100 GPU，售价12.9万美元。黄仁勋亲自把第一台DGX-1装进他的车里，开车送到旧金山交给OpenAI。这是NVIDIA与AI研究社区深度绑定的标志性事件——OpenAI后来成为ChatGPT的缔造者，ChatGPT又引爆了AI算力需求的指数级增长，形成了一个完美的正循环[s17]。

2018年，英伟达推出RTX系列，全球首款支持实时光线追踪（Ray Tracing）的GPU。光线追踪是好莱坞电影特效的标准技术，但此前只能在电影渲染农场中离线计算。RTX将它带到了实时渲染领域，彻底重新定义了计算机图形学。数据中心业务开始快速增长，AI转型全面加速[s7]。

2022年成为英伟达历史上最关键的一年。2月，公司宣布放弃400亿美元收购Arm的交易——这笔交易原本将创造半导体史上最大的收购案，但因监管阻力最终流产，英伟达支付12.5亿美元分手费。这笔"分手费"看似巨大，但对于一个即将起飞的公司来说微不足道。11月，ChatGPT发布，大模型训练需求爆发。H100 GPU供不应求，交货周期长达数月。数据中心收入开始指数级增长[s11]。

从FY2023的约150亿美元，到FY2024的475亿美元，到FY2025的约1,150亿美元，到FY2026达到1,937亿美元——数据中心收入的年复合增长率超过100%。这不仅是增长，这是指数级爆发。

市值的狂飙令人瞠目：2023年5月，英伟达市值首次突破1万亿美元，成为首家市值破万亿的芯片公司，全年增长240%。2024年3月，Blackwell架构发布（B200），集成2,080亿个晶体管（双die设计，每die 1,040亿），第五代Tensor Core，原生FP4支持，推理性能较H100提升最高30倍。6月，市值突破3万亿美元，一度超越微软成为全球市值最高公司。黄仁勋入选《时代》周刊100位最具影响力人物（第二次），当选美国工程院院士[s11]。

2025年是英伟达创造历史的一年。7月市值突破4万亿美元；10月29日，成为历史上首家收盘市值突破5万亿美元的公司。黄仁勋以"AI的缔造者"身份入选《时代》周刊2025年度人物，位列《财富》杂志全球100位最具影响力商界人士第一位。11月，他在剑桥大学演讲获Stephen Hawking Fellowship，说"to be a CEO is a lifetime of sacrifice"[s11][s21]。

2026年，英伟达的步伐进一步加速。1月CES发布GeForce RTX 50系列和Cosmos基础模型。3月GTC 2026发布Vera Rubin下一代架构，黄仁勋称其为"NVIDIA公司史上最雄心勃勃的事业"，动用全公司40,000名工程师。Vera Rubin GPU搭载288GB HBM4显存，Vera CPU采用88核Arm架构，性能达x86的1.8倍。5月13日，市值盘中突破5.5万亿美元，超越德国全年GDP。8月27日，FY2027 Q2财报显示营收962亿美元（同比+106%），黄仁勋宣告"AI已到达转折点"。6月GTC Taipei演讲近20分钟，发布RTX Spark超级芯片，宣布Vera Rubin全面量产，中国台湾150多家供应链合作伙伴参与[s7][s11]。

从1万亿到5万亿，英伟达仅用了两年半时间，创造了商业史上前所未有的增长奇迹。而这一切的起点，是2012年两块500美元的显卡和一个名叫AlexNet的神经网络[s12][s13]。

**English:** In September 2012, in a bedroom at his parents' house, Alex Krizhevsky, a PhD student at the University of Toronto, spent five to six days training a deep convolutional neural network on two GeForce GTX 580 gaming cards (about US$500 each). It was called AlexNet. His co-authors were Ilya Sutskever and his supervisor Geoffrey Hinton, one of the three giants of deep learning. AlexNet achieved a top-5 error rate of 15.3 percent in the ImageNet competition, against 26.2 percent for the runner-up—a gap of more than ten percentage points, not an incremental improvement but a landslide breakthrough. Huang called it "the Big Bang of AI"[s12][s13].

NVIDIA grasped the significance faster than almost anyone. The company had spent six years watching GPU computing spread through science and high-performance computing—molecular dynamics, climate simulation, fluid mechanics. But AlexNet was different. It hinted that the most exciting application of parallel computing might not be physics but intelligence itself: not simulating the behaviour of the universe, but simulating the most complex structure in the universe—the brain. NVIDIA quickly reoriented the company: the Tesla product line was repositioned as a data-centre computing line, and successive architectures (Kepler, Maxwell, Pascal, Volta, Turing, Ampere, Hopper, Blackwell, Rubin) were increasingly optimised for deep-learning workloads[s12].

In 2016 NVIDIA unveiled two landmark products and initiatives. The first was the Pascal architecture (P100), the first GPU designed specifically for deep learning, alongside the debut of the NVLink high-speed interconnect—meaning communication speed between GPUs would no longer be the bottleneck in training. The second was the DGX-1, "a supercomputer in a box," packing eight Tesla P100 GPUs and priced at US$129,000. Huang personally loaded the first DGX-1 into his car and drove it to San Francisco to hand over to OpenAI. It was the signature event in NVIDIA's deep binding to the AI research community: OpenAI later built ChatGPT, and ChatGPT in turn detonated exponential demand for AI compute, forming a perfect virtuous circle[s17].

In 2018 NVIDIA launched the RTX line, the world's first GPUs to support real-time ray tracing. Ray tracing was the standard technique for Hollywood special effects, but until then it could only be computed offline in movie render farms. RTX brought it to real-time rendering, redefining computer graphics once again. The data-centre business began to grow rapidly, and the AI transformation accelerated across the board[s7].

2022 became the most pivotal year in NVIDIA's history. In February the company announced it was abandoning the US$40 billion acquisition of Arm—a deal that would have been the largest in semiconductor history, but which collapsed under regulatory resistance, with NVIDIA paying a US$1.25 billion break-up fee. The "break-up fee" looked large but was trivial for a company about to take off. In November, ChatGPT was released and demand for large-model training exploded. The H100 GPU could not be made fast enough, with lead times stretching to months. Data-centre revenue began its exponential climb[s11].

From roughly US$15 billion in FY2023, to US$47.5 billion in FY2024, to about US$115 billion in FY2025, to US$193.7 billion in FY2026—data-centre revenue grew at a compound annual rate exceeding 100 percent. This was not mere growth; it was exponential detonation.

The market-value surge was staggering: in May 2023 NVIDIA's capitalisation crossed US$1 trillion for the first time, making it the first chip company to reach the milestone; the stock rose 240 percent over the year. In March 2024 the Blackwell architecture (B200) was launched, integrating 208 billion transistors (a dual-die design, 104 billion per die), with fifth-generation Tensor Cores, native FP4 support and up to 30 times the inference performance of the H100. In June the market value broke US$3 trillion, briefly overtaking Microsoft to become the world's most valuable company. Huang was named to Time's list of the 100 most influential people (for the second time) and elected to the US National Academy of Engineering[s11].

2025 was a year in which NVIDIA made history. In July its market value surpassed US$4 trillion; on October 29 it became the first company in history to close with a market value above US$5 trillion. Huang was named Time's 2025 Person of the Year as "the builder of AI" and ranked first on Fortune's list of the world's 100 most influential business leaders. In November, receiving the Stephen Hawking Fellowship with a lecture at the University of Cambridge, he said that "to be a CEO is a lifetime of sacrifice"[s11][s21].

In 2026 NVIDIA's pace accelerated further. In January, at CES, it launched the GeForce RTX 50 series and the Cosmos foundation models. In March, at GTC 2026, it unveiled the next-generation Vera Rubin architecture, which Huang called "the most ambitious undertaking in the company's history," mobilising all 40,000 of the company's engineers. The Vera Rubin GPU carries 288GB of HBM4 memory; the Vera CPU uses an 88-core Arm architecture with 1.8 times the performance of x86. On May 13 the market value broke US$5.5 trillion intraday, exceeding Germany's entire annual GDP. On August 27, FY2027 Q2 results showed revenue of US$96.2 billion (up 106 percent year on year), and Huang declared that "AI has reached its tipping point." His nearly twenty-minute GTC Taipei keynote in June introduced the RTX Spark superchip, announced that Vera Rubin was in full volume production, and noted the participation of more than 150 supply-chain partners in Taiwan[s7][s11].

From US$1 trillion to US$5 trillion took NVIDIA only two and a half years—a growth miracle without precedent in business history. And the starting point of it all was two US$500 graphics cards and a neural network called AlexNet in 2012[s12][s13].

### 8. 管理哲学：痛苦、扁平与极致透明 / Management Philosophy: Pain, Flatness and Radical Transparency

黄仁勋的管理风格在科技界独树一帜，既令人敬畏又令人着迷。它融合了工程师的精确、移民的饥饿感和肯塔基山区男孩的坚韧，形成了一种独特的"英伟达式"管理哲学。他的管理哲学可以追溯到肯塔基那段扫厕所的日子——"伟大来自品格，而品格不是聪明人天生的，是苦过的人磨出来的"[s15][s16]。

他经常说"伟大需要大量的痛苦和磨难"（Greatness requires ample doses of pain and suffering）。他自称"firing-averse"（不喜欢开除人），但喜欢"折磨人到伟大"（torture people into greatness）。这句话半开玩笑半认真——他的方式是把人推到极限，但目的是让他们成长，而不是让他们离开。在Stripe活动中他说："我不喜欢放弃人。我宁愿折磨他们直到变得伟大。"[s23]

新书《思考机器》（The Thinking Machine）作者Stephen Witt花了6个小时深度采访黄仁勋后得出一个惊人结论：黄仁勋几乎完全被负面情绪驱动——"他的动力很大程度上来自恐惧和内疚：对失败的恐惧、对竞争的偏执，以及对让别人失望的愧疚。"这颠覆了人们对成功CEO的传统想象——他们通常被描述为充满激情的愿景型领导者。但黄仁勋不一样。他凌晨4点准时睁眼，望着天花板思考英伟达可能失败的各种方式，然后起床工作。"我不想失败的动力远大于想成功的动力"——这与传统认知中的"乐观进取型"CEO截然不同，更像是一个从底层爬上来的人特有的生存焦虑[s20][s22]。

在组织结构上，黄仁勋维持着极为扁平的管理模式。他直接管理约50-60名下属（Fortune报道为60个，Lex Fridman播客中他也确认是60个），这在大型科技公司中极为罕见——标准的CEO通常只有8-12个直接下属。为什么这么做？在Lex Fridman播客中，他解释了这个设计的逻辑："公司的架构应该反映它所生产的产品。"英伟达的产品（AI超级计算机）是极端协同设计的产物——内存、光学、散热、网络、软件必须同时优化，不能各自为战。因此，他的组织也必须让所有领域的专家能够同时看到问题、同时讨论。如果一个专家提出的散热方案影响了配电，其他专家可以立刻发现问题并当场讨论解决方案。这种"所有人同时看到所有问题"的模式，与传统的层层汇报、逐级审批的模式完全相反[s20][s25]。

黄仁勋不做一对一会议。重要讨论他喜欢在团队会议中公开进行。他的理由是："所有英伟达高管都应该能够从我给任何一个人的反馈中学习，他们都应该从观察我一起解决问题的过程中受益。"项目负责人可能突然发现自己直接向黄仁勋汇报，这些人被称为"pilots in charge"（主管飞行员）——这个航空比喻暗示了一种精英主义：你是飞行员，不是乘客[s25]。

他推行"五点邮件制度"（T5T，Top 5 Things）。据巴伦周刊记者金泰（Tae Kim）在《The Nvidia Way》一书中披露，全公司约3万名员工每周都要发送"Top 5"邮件，用五点总结最重要的工作进展。黄仁勋每天阅读大量T5T邮件，从中获取一线的"弱信号"（weak signals）。他说："捕捉强信号很容易，但我想在它们还弱的时候就截获。"这是他对抗大公司官僚病的方法——当公司逐渐庞大后，坏消息到不了CEO耳朵里，CEO被隔绝在真相之外，这是每个大公司都面临的"信息衰减"问题。T5T邮件让他能看到基层员工注意到的、高层可能还毫无察觉的趋势。腾讯科技专访中，《黄仁勋：英伟达之芯》作者Stephen Witt描述："他要求公司全体员工每周五给他发邮件，列出最重要的五项工作，然后随机选取一些邮件查阅，有时候甚至还会进行回复。一个基层员工，给一家3万人的大公司CEO写一封电子邮件，然后几分钟后他就会回复——这在其他公司是不存在的。"[s23][s25][s38]

黄仁勋不相信私下批评。当出现问题时，他会在公开场合讨论——一个人可能当场难堪，但所有人都能学到教训。据报道，"在英伟达，你要习惯站在一个有100人的房间里，直面老板对你持续30分钟的大喊大叫。"他会特意等到有"观众"的会议场合再批评，让公开批评变成教学示范——一个人犯错，所有人都受益。但员工们也承认，他从不会无缘无故发脾气——每一次批评都是针对具体问题，逻辑清晰，有理有据，目的是让所有人都学到教训。他认为"从自己的尴尬中学习有用，从别人的错误中学习更高效"[s15][s23]。

尽管要求严苛，英伟达的员工流失率却极低。FY2025整体流失率据报道仅约2.5%——在硅谷科技公司的普遍20-30%流失率面前，这是一个惊人的数字。约五分之一的员工在公司超过十年。黄仁勋将这归因于"使命是老板"（The mission is the boss）的文化——员工服务于使命而非经理，这避免了大公司常见的政治内斗和决策迟缓。从游戏转向机器学习的战略调整，仅通过一封全公司周五晚间邮件就完成了传达[s23][s25]。

他每天工作12-14小时，每周7天。他说自己"不要无聊，也不要被解雇"——这是他对"世界上任职时间最长的科技CEO"秘诀的总结。在剑桥大学的Stephen Hawking Fellowship演讲上，他说："当CEO是一生的牺牲。大多数人以为是领导、是指挥、是站在顶峰。都不是。你是在为公司服务。"[s21]

**English:** Jensen Huang's management style stands alone in the technology world, at once feared and fascinating. It fuses an engineer's precision, an immigrant's hunger and the toughness of a Kentucky mountain boy into a distinctive "NVIDIA-style" management philosophy. The philosophy traces back to his days cleaning toilets in Kentucky: "Greatness comes from character, and character is not something smart people are born with—it is forged by those who have suffered"[s15][s16].

He often says that "greatness requires ample doses of pain and suffering." He describes himself as "firing-averse" but fond of "tortur[ing] people into greatness." The line is half joke, half earnest: his method is to push people to their limits, but the aim is to make them grow, not to make them leave. At a Stripe event he said: "I don't like giving up on people. I would rather torture them until they become great."[s23]

Stephen Witt, author of the new book The Thinking Machine, spent six hours interviewing Huang in depth and reached a striking conclusion: Huang is driven almost entirely by negative emotions—"his motivation comes largely from fear and guilt: fear of failure, paranoia about competition, and guilt about letting others down." That upends the conventional image of the successful CEO, usually portrayed as a passionate, visionary optimist. Huang is different. He opens his eyes promptly at four in the morning, stares at the ceiling thinking through the ways NVIDIA might fail, then gets up and works. "My motivation not to fail is far greater than my motivation to succeed"—a far cry from the conventionally "optimistic, driven" CEO, and much closer to the survival anxiety peculiar to a man who climbed up from the bottom[s20][s22].

In organisational structure Huang maintains an extraordinarily flat management model. He directly manages about 50 to 60 subordinates (Fortune reports 60, a figure he himself confirmed on the Lex Fridman podcast), which is extremely rare at a large technology company—a standard CEO usually has only 8 to 12 direct reports. Why? On the Lex Fridman podcast he explained the logic of the design: "The architecture of the company should reflect the products it makes." NVIDIA's products—AI supercomputers—are the product of extreme co-design: memory, optics, cooling, networking and software must all be optimised together, not separately. So his organisation, too, must let experts in every field see the problems and discuss them at the same time. If a cooling solution proposed by one expert affects power distribution, the other experts can spot it immediately and thrash out a solution on the spot. This model of "everyone seeing every problem at once" is the opposite of the traditional pattern of layered reporting and sequential approval[s20][s25].

Huang does not hold one-on-one meetings. He prefers important discussions to happen openly in team meetings. His reasoning: "Every NVIDIA executive should be able to learn from the feedback I give to any single person, and they should all benefit from watching me solve problems together with them." Project leads may suddenly find themselves reporting directly to Huang; these people are called "pilots in charge"—an aviation metaphor implying a certain elitism: you are the pilot, not a passenger[s25].

He runs the "Top 5 Things" email system (T5T). According to Barron's reporter Tae Kim, in his book The Nvidia Way, all of the company's roughly 30,000 employees send a "Top 5" email each week, summing their most important progress in five points. Huang reads large numbers of T5T emails every day to glean front-line "weak signals." He says: "It's easy to catch strong signals, but I want to intercept them while they are still weak." It is his method for fighting the bureaucratic disease of large companies: as a company grows, bad news stops reaching the CEO, who becomes sealed off from the truth—the "information decay" every large company faces. T5T emails let him see trends that rank-and-file employees notice and senior leaders may be completely unaware of. In a Tencent Tech interview, Stephen Witt, author of Jensen Huang: The NVIDIA Core, described it: "He asks every employee of the company to email him every Friday listing their five most important items of work, and then he randomly selects some of the emails to read, sometimes even replying. A junior employee writes an email to the CEO of a 30,000-person company, and minutes later he writes back—that simply doesn't exist at other companies."[s23][s25][s38]

Huang does not believe in private criticism. When something goes wrong, he discusses it in public; one person may be embarrassed on the spot, but everyone learns the lesson. Reportedly, "at NVIDIA, you have to get used to standing in a room of 100 people and facing your boss yelling at you continuously for thirty minutes." He deliberately waits for meetings with an "audience" before criticising, turning public criticism into a teaching demonstration: one person makes the mistake, everyone benefits. Employees also concede that he never loses his temper without cause—each criticism targets a concrete problem, logically clear and well-founded, aimed at letting everyone learn. In his view, "learning from your own embarrassment is useful, but learning from other people's mistakes is more efficient"[s15][s23].

For all the rigour, NVIDIA's attrition is remarkably low. The overall attrition rate in FY2025 was reported at only about 2.5 percent—an astonishing figure against the 20 to 30 percent typical of Silicon Valley technology companies. Roughly one in five employees has been at the company for more than a decade. Huang attributes this to the culture of "the mission is the boss": employees serve the mission rather than a manager, which avoids the political infighting and slow decision-making common in large companies. The strategic pivot from gaming to machine learning was communicated in a single company-wide Friday-night email[s23][s25].

He works 12 to 14 hours a day, seven days a week. His formula for being, as he puts it, the longest-serving technology CEO in the world: "Don't be bored, and don't get fired." Delivering the Stephen Hawking Fellowship lecture at Cambridge, he said: "To be a CEO is a lifetime of sacrifice. Most people think it's about leading, about commanding, about standing at the summit. It's none of those things. You are serving the company."[s21]

### 9. 个人生活与公众形象：皮衣、家庭与中国情结 / Private Life and Public Image: The Leather Jacket, Family and Chinese Roots

黄仁勋的个人生活相对低调，但近年来他的公众形象已超越了一般科技CEO的范畴，成为一种文化现象——一种融合了技术天才、移民传奇和极简美学的独特符号。

1985年，黄仁勋与大学实验课搭档Lori Mills结婚。两人的相识源于俄勒冈州立大学的电气工程系——当时250名学生中只有3位女生，Lori是其中之一。她不仅是他的同学，更是他工程实验课上的搭档（lab partner）——在实验室里并肩工作的经历，培养了两人之间独特的默契和信任。两人1985年结婚，至今已携手走过四十年。他们育有一子一女：长子黄盛斌（Spencer Huang），现任英伟达产品经理；小女儿黄敏珊（Madison Huang），任产品营销总监。值得一提的是，Spencer曾在中国台湾台北开设酒吧，入选亚洲Top 50酒吧榜单（2015-2021年），这段创业经历颇有乃父之风——在Denny's打工的黄仁勋，其子也在酒吧行业闯荡过。2021年酒吧关闭后，Spencer加入了英伟达[s1][s2]。

黄仁勋能说流利的英语、普通话和闽南语（台语）。他在AMD工作时开始系统学普通话，目的是和中国光罩工人沟通——这种从基层开始学习语言的方式，体现了他一贯的务实风格。每年回中国台湾参加Computex时，他都会用普通话和闽南语与媒体和供应链伙伴交流，这种语言能力让他在华人商业圈中如鱼得水[s1]。

自2013年前后起，黄仁勋开始以黑色皮衣、黑衬衫搭配牛仔裤的造型出现在公众场合，被中国网友亲切地称为"皮衣黄"。这一标志性造型已成为他个人品牌的核心符号——就像乔布斯的黑色高领衫、扎克伯格的灰色T恤一样，成为科技领袖"极简主义"审美的又一个经典案例。皮衣品牌从未公开披露（网上流传的Tom Ford说法仅为推测，未获任何官方或可靠来源证实）。在韩国访问时，"Jensen Huang jacket"搜索量激增130%。Facebook创始人Mark Zuckerberg曾说他是"科技界的Taylor Swift"——形容他在年轻人心中的超级偶像级影响力[s16]。

2024年Computex期间，中国台湾媒体创造了"Jensanity"一词来形容围绕黄仁勋的狂热现象——粉丝们排队数小时只为见他一面、摸一摸他的皮衣、听他的一场演讲。这种个人魅力和粉丝文化在科技CEO中极为罕见，已经超越了商业范畴，成为一种社会文化现象[s16]。

黄仁勋每年回中国台湾参加Computex，与中国台湾供应链合作伙伴保持密切关系。在2026年GTC Taipei/Computex演讲中，他宣布Vera Rubin已进入全面量产，中国台湾150多家供应链合作伙伴参与，供应链规模是Grace Blackwell的两倍——这表明中国台湾在全球AI芯片供应链中的地位不降反升[s11][s32]。

他与AMD CEO苏姿丰（Lisa Su）的亲戚关系也是公众津津乐道的话题。两人的关系是：黄仁勋的母亲是苏姿丰外祖父的幼妹，即表舅甥关系（first cousins once removed）。在芯片行业，两大家族分别执掌着最大的两家GPU公司，既是亲戚又是竞争对手——这种戏剧性的关系为科技行业增添了一抹人情味。苏姿丰本人已确认这是"distant relative"关系[s1]。

在慈善方面，黄仁勋设立了黄氏基金会（Jen-Hsun & Lori Huang Foundation），但具体捐赠规模和方式在公开信息中较为有限。2019年，他向肯塔基州的Oneida Baptist Institute捐赠200万美元修建女生宿舍——这不是出于义务，而是出于感恩。此外，他拥有多所大学的荣誉博士学位，包括中国台湾阳明交通大学、中国台湾大学、俄勒冈州立大学、华中科技大学和瑞典林雪平大学[s1][s3]。

黄仁勋的净资产几乎全部与英伟达股价挂钩——他持有约3-3.5%的英伟达股份。净资产数字随股价大幅波动：2025年11月Bloomberg Billionaires Index约为1,650亿美元，2026年8月约为1,968亿美元（Forbes Argentina数据，非福布斯主站数据）。财富排名的变化本身就是英伟达股价波动的缩影。他的个人生活并没有因为财富的暴涨而发生根本性改变——他依然每天凌晨4点起床，依然穿那件黑色皮衣，依然自己开车（至少在某些场合）[s20][s26]。

**English:** Jensen Huang's private life is relatively low-key, yet in recent years his public image has transcended that of the ordinary technology CEO to become a cultural phenomenon—a singular symbol fusing technical genius, immigrant saga and minimalist aesthetics.

In 1985 Huang married Lori Mills, his lab partner from university. The two met in the electrical-engineering department at Oregon State University—of the 250 students then enrolled, only three were women, and Lori was one. She was not merely a classmate but his lab partner in engineering experiments; working side by side in the laboratory forged a distinctive rapport and trust between them. They married in 1985 and have now been together for forty years. They have a son and a daughter: their elder son, Spencer Huang, is now a product manager at NVIDIA; their younger daughter, Madison Huang, is a director of product marketing. Notably, Spencer once ran a bar in Taipei, Taiwan, that made Asia's Top 50 Bars list (2015-2021)—an entrepreneurial venture very much in his father's spirit: the Jensen Huang who worked at Denny's had a son who tried his luck in the bar trade. After the bar closed in 2021, Spencer joined NVIDIA[s1][s2].

Huang speaks fluent English, Mandarin and Hokkien (Taiwanese). He began systematically learning Mandarin while at AMD, in order to communicate with Chinese mask workers—learning a language from the ground up, in characteristically pragmatic fashion. Each year, returning to Taiwan for Computex, he converses with media and supply-chain partners in Mandarin and Hokkien, a fluency that lets him move through the Chinese-speaking business world with ease[s1].

From around 2013 onward, Huang began appearing in public in a black leather jacket, black shirt and jeans, affectionately dubbed "Leather Jacket Huang" by Chinese internet users. The signature look has become the core symbol of his personal brand—alongside Steve Jobs's black turtleneck and Mark Zuckerberg's grey T-shirt, another classic of the technology leader's "minimalist" aesthetic. The jacket's brand has never been publicly disclosed (the Tom Ford claim circulating online is mere speculation, unconfirmed by any official or reliable source). During a visit to South Korea, searches for "Jensen Huang jacket" surged 130 percent. Facebook founder Mark Zuckerberg has called him "the Taylor Swift of the tech world"—describing his idol-level influence among the young[s16].

During Computex 2024, Taiwanese media coined the word "Jensanity" to describe the frenzy around Huang—fans queuing for hours just to catch a glimpse of him, touch his leather jacket or hear one of his talks. Such personal charisma and fan culture, exceedingly rare among technology CEOs, has moved beyond commerce to become a social and cultural phenomenon[s16].

Huang returns to Taiwan every year for Computex and maintains close ties with Taiwanese supply-chain partners. In his 2026 GTC Taipei/Computex keynote, he announced that Vera Rubin had entered full volume production with the participation of more than 150 supply-chain partners in Taiwan, the supply-chain footprint being twice that of Grace Blackwell—evidence that Taiwan's position in the global AI-chip supply chain has strengthened rather than weakened[s11][s32].

His family tie to AMD CEO Lisa Su is also a topic of public fascination. The relationship: Huang's mother was the youngest sister of Su's maternal grandfather, making them first cousins once removed. In the chip industry, the two families respectively run the two largest GPU companies—relatives and rivals at once, a dramatic relationship that adds a touch of human warmth to the technology business. Su herself has confirmed they are "distant relatives"[s1].

In philanthropy, Huang established the Jen-Hsun & Lori Huang Foundation, though the scale and methods of its giving are sparsely documented in public information. In 2019 he donated US$2 million to the Oneida Baptist Institute in Kentucky to build a girls' dormitory—not out of obligation but out of gratitude. He also holds honorary doctorates from several universities, including National Yang Ming Chiao Tung University in Taiwan, National Taiwan University, Oregon State University, Huazhong University of Science and Technology and Linkoping University in Sweden[s1][s3].

Huang's net worth is almost entirely tied to NVIDIA's share price—he holds roughly 3 to 3.5 percent of the company. The figure swings with the stock: around US$165 billion on the Bloomberg Billionaires Index in November 2025, and about US$196.8 billion in August 2026 (per Forbes Argentina data, not the main Forbes site). The shifts in his wealth ranking are themselves a miniature of NVIDIA's share-price volatility. His personal life has not fundamentally changed with the explosion of wealth: he still rises at four in the morning, still wears the black leather jacket, still drives himself (at least on some occasions)[s20][s26].

### 10. 中国市场与出口管制：从95%到接近归零 / The China Market and Export Controls: From 95 Percent to Near Zero

中国市场对英伟达的意义非同一般。作为全球领先的半导体消费市场之一，中国曾贡献了英伟达收入的相当大比例。黄仁勋深知中国市场的重要性——他不仅从商业角度理解中国，更从文化、语言和家族渊源的角度与中国有着天然的联系。然而，近年来美国出口管制的层层升级，彻底改变了英伟达在中国的商业格局，也让他陷入了一种罕见的困境——作为全球最成功的AI芯片公司的CEO，他最大的市场之一正在对他关闭大门[s27]。

2022年10月，美国发布首轮对华芯片出口管制规定，限制向中国出口先进AI芯片和半导体制造设备。这一管制直接瞄准了AI训练所需的高端GPU——正是英伟达最核心的产品。2023年10月，管制进一步升级，收紧了对算力性能和芯片互联带宽的限制，堵住了此前的一些"灰色地带"。英伟达为中国市场专门开发了H20特供芯片，其算力约为H100的六分之一——这是在合规框架内能提供的最高端产品，但对于训练大模型来说远远不够[s27]。

然而管制的收紧并未停止。2025年4月，H20芯片的出口许可证被要求暂停——这意味着连"阉割版"也不能卖了。2025年7月，H20恢复销售。10月，黄仁勋首次公开承认受出口管制影响，英伟达在中国高端AI芯片市场的份额从高峰期的95%骤降至接近零——从"几乎垄断"到"几乎消失"，只用了三年时间。12月，特朗普政府批准H200对华出口，但附加了一个前所未有的条件：英伟达需将25%的销售收入上缴美国财政部，并接受安全审查[s27][s28]。

2026年1月，美国工业和安全局（BIS）将H200出口改为逐案审查模式——这意味着每一笔交易都需要单独审批，不确定性大大增加。3月GTC大会上，黄仁勋宣布已获中国客户H200订单、重启H200生产。下半年开始有少量H200交付到字节跳动、腾讯等公司，但总量严重受限，且还需中国发改委审批——出口管制不仅限制了卖方的出口，也限制了买方的进口[s27][s28]。

与此同时，中国国产替代加速推进。黄仁勋本人在Bloomberg采访中承认，华为昇腾等国产芯片已达到H200级别性能。据市场预计，国产芯片2026年将占据近90%的中国高端AI芯片市场——这是一个令人震惊的数字。英伟达在中国市场面临的挑战是双重的——既受美国出口管制限制，又面对中国本土竞争对手的快速崛起[s27]。

黄仁勋一直公开反对出口管制。2025年7月链博会演讲中，他说："任何想出这步棋、认为禁止H20芯片出口就能阻止中国发展AI的人，都是非常无知的。"2026年8月，他在与戴尔同台的Bloomberg访谈中警告："我们并不是在整体上远远领先于中国"（"We're not far ahead of China overall"）——这番坦率的表态在美国引起了争议，但黄仁勋选择说出事实而非政治正确的话[s27][s29]。

在2026年3月的GTC大会上，黄仁勋一方面宣布重启H200对华生产，另一方面也在加速全球布局。他在不同场合反复强调一个观点：AI基础设施的竞争是全球性的，各国都在竞相建设"AI工厂"，而英伟达的目标是成为这场竞赛的核心供应商。他提出了"Token经济学"的概念——"Token现在是盈利的单位。每一个Token都是收入。数据中心不再是存储设施，而是Token生产工厂。"在这个框架下，英伟达不是在卖芯片，而是在为全球的AI基础设施建设"发电厂"[s11][s14]。

需要说明的是，出口管制与芯片贸易政策仍在快速调整中。本章所述动态截至2026年9月，后续审批规则、交付进展与市场份额变化仍需持续追踪——作为一份长期存证的档案，我们在此标注时间边界，以免后来的读者以静态文本误读动态局势。

**English:** The China market means a great deal to NVIDIA. As one of the world's largest markets for semiconductor consumption, China once contributed a substantial share of NVIDIA's revenue. Huang understands the importance of China deeply—he grasps the country not only from a commercial standpoint but through natural ties of culture, language and family history. Yet the successive escalation of US export controls in recent years has utterly reshaped NVIDIA's commercial position in China, landing him in a rare predicament: as CEO of the world's most successful AI-chip company, one of his largest markets is closing its doors to him[s27].

In October 2022, the United States issued its first round of chip export controls on China, restricting the export of advanced AI chips and semiconductor-manufacturing equipment. The controls took direct aim at the high-end GPUs required for AI training—precisely NVIDIA's most core products. In October 2023 the controls were tightened further, narrowing the limits on computing performance and chip-to-chip interconnect bandwidth and closing off several earlier "grey areas." NVIDIA developed the H20 specifically for the Chinese market, a chip with roughly one-sixth the computing power of the H100—the most advanced product permissible within the compliance framework, yet far from sufficient for training large models[s27].

The tightening did not stop there. In April 2025, export licences for the H20 were ordered suspended—meaning even the "castrated" version could no longer be sold. In July 2025 H20 sales resumed. In October, Huang publicly acknowledged for the first time that, under the impact of export controls, NVIDIA's share of China's high-end AI-chip market had plummeted from a peak of 95 percent to nearly zero—from "near monopoly" to "near disappearance" in just three years. In December, the Trump administration approved H200 exports to China, but attached an unprecedented condition: NVIDIA must hand over 25 percent of the sales revenue to the US Treasury and submit to security review[s27][s28].

In January 2026, the US Bureau of Industry and Security (BIS) switched H200 exports to a case-by-case review model—meaning every transaction requires separate approval, greatly increasing uncertainty. At the GTC conference in March, Huang announced that H200 orders from Chinese customers had been received and H200 production restarted. Small quantities of H200s began reaching companies such as ByteDance and Tencent in the second half of the year, but total volumes were severely limited and still required approval from China's National Development and Reform Commission—the export controls restrict not only the seller's exports but the buyer's imports as well[s27][s28].

Meanwhile, China's domestic substitution is accelerating. Huang himself admitted in a Bloomberg interview that home-grown chips such as Huawei's Ascend had reached H200-class performance. By market estimates, domestic chips will take nearly 90 percent of China's high-end AI-chip market in 2026—a startling figure. NVIDIA faces a double challenge in China: constrained both by US export controls and by the rapid rise of indigenous Chinese competitors[s27].

Huang has consistently and publicly opposed export controls. In a July 2025 speech at the China International Supply Chain Expo, he said: "Anyone who came up with this move, anyone who thinks that banning H20 chip exports will stop China from developing AI, is extremely naive." In August 2026, in a Bloomberg interview alongside Dell, he warned: "We're not far ahead of China overall"—a candid statement that stirred controversy in the United States, but Huang chose to state facts rather than politically correct platitudes[s27][s29].

At GTC in March 2026, even as he announced the restart of H200 production for China, Huang was accelerating the global layout. In various settings he has repeatedly stressed one point: the competition in AI infrastructure is global, with every country racing to build "AI factories," and NVIDIA's aim is to be the central supplier in that race. He has put forward the concept of the "token economy": "Tokens are now the unit of profitability. Every token is revenue. Data centres are no longer storage facilities but token-production factories." In that framework NVIDIA is not selling chips but building "power plants" for the world's AI infrastructure[s11][s14].

It should be noted that export controls and chip-trade policy remain in rapid flux. The developments described in this chapter are current as of September 2026; subsequent approval rules, delivery progress and shifts in market share will require continued tracking. As an archive kept for long-term record, we mark this time boundary here, lest later readers misread a dynamic situation through static text.

### 11. AGI宣言与未来：从Vera Rubin到物理AI / The AGI Declaration and the Future: From Vera Rubin to Physical AI

2026年3月22日，在Lex Fridman播客第494期中，黄仁勋说了一句令全球科技界震动的话："我认为就是现在。我认为我们已经实现了AGI。"（"I think it's now. I think we've achieved AGI."）[s14]

这番表态的背景值得深入分析。首先，需要理解黄仁勋的AGI定义——他的标准是"经济性的"而非学术性的：一个能自主创造价值10亿美元服务的AI。这与学术界传统定义中的"在所有认知任务上达到人类水平"有本质区别。学术界的AGI意味着一个能像人类一样思考、学习和创造的智能体——而黄仁勋的AGI则更像一个"经济代理人"：它不需要在所有方面都像人，只要能独立创造足够的经济价值。其次，他同时进行了自我修正——当Lex Fridman追问"AI能复制英伟达这样复杂的公司吗？"时，黄仁勋直接回答："概率为零"（"The probability is zero"）。10万个AI智能体也无法重建英伟达。这个自我修正非常关键——它表明黄仁勋对AI能力有着清醒的认知[s14]。

这番表态的时机选择耐人寻味。当时正值GTC 2026期间，英伟达宣布了至少1万亿美元的Blackwell+Rubin订单。市场反应积极：NVDA股价上涨1.7%，AI相关加密货币跳涨10-20%，Polymarket上"2027年AGI实现"概率从15%飙升至40%。但绝大多数AI研究者认为当前系统远未达到真正的AGI。Meta首席AI科学家Yann LeCun等学者多次质疑商业领袖们在"移动球门柱"——每当AI接近某个目标时，就重新定义目标[s14][s40]。

黄仁勋对未来技术的布局可以从三个维度理解，每个维度都指向同一个核心判断——AI正在从数字世界向物理世界渗透，而英伟达要在两个世界都建立不可替代的地位。

第一，算力基础设施的代际跃迁——Blackwell与Vera Rubin。Blackwell架构（2024年3月发布）拥有2,080亿个晶体管（双die设计），第五代Tensor Core，原生FP4支持，推理性能较H100提升最高30倍。Vera Rubin（2026年3月发布）更进一步，动用全公司40,000名工程师开发，是"NVIDIA公司史上最雄心勃勃的事业"。Vera Rubin GPU搭载288GB HBM4显存，Vera CPU采用88核Arm架构，性能达x86的1.8倍。黄仁勋宣布Vera Rubin已进入全面量产，中国台湾供应链规模是上一代的两倍。从Blackwell到Vera Rubin，英伟达保持着每两年一代架构的惊人节奏[s7][s11][s39]。

第二，物理AI——下一个ChatGPT时刻。黄仁勋在GTC 2026上宣告"物理AI的ChatGPT时刻即将到来"，发布了Cosmos 3物理AI平台，押注机器人与物理世界AI。他认为AI不仅存在于数字世界（语言模型、图像生成），更将深入物理世界——工厂、仓库、汽车、机器人都需要AI来感知、理解和操作物理环境。Cosmos平台旨在为机器人提供"物理世界理解"的能力——这将是AI的下一个巨大 Frontier[s14]。

第三，Token经济学——AI时代的新经济模型。黄仁勋提出了一个全新的经济框架："Token是AI时代的新货币。"在他的叙述中，每一个AI生成的Token（一个词、一个像素、一个决策）都是有经济价值的——它们是"收入"的微观单元。数据中心从"存储设施"变成了"Token生产工厂"。全球正在竞相建设AI工厂，这是"人类历史上规模最大的基础设施建设"。在这个框架下，英伟达不是卖芯片的公司，而是为AI经济提供"发电设备"的公司[s14]。

2026年5月13日，英伟达市值盘中突破5.5万亿美元，超越德国全年GDP——一个单一公司的市值超过了一个主要经济体的全年经济产出，这在人类商业史上是前所未有的。8月27日，FY2027 Q2财报显示营收962亿美元（同比+106%），黄仁勋宣告"AI已到达转折点"。英伟达的故事还远未结束——正如他常说的："我们距离倒闭永远只有30天。"[s11][s19]

**English:** On March 22, 2026, on episode 494 of the Lex Fridman podcast, Jensen Huang said something that jolted the global technology world: "I think it's now. I think we've achieved AGI."[s14]

The background to the statement repays close analysis. First, one must understand Huang's definition of AGI: his standard is "economic" rather than academic—an AI capable of autonomously creating a service worth US$1 billion. That differs fundamentally from the academic convention of "reaching human-level performance on all cognitive tasks." The academic AGI means an agent that can think, learn and create like a human being; Huang's AGI is more like an "economic agent"—it need not resemble humans in every respect, so long as it can independently create sufficient economic value. Second, he immediately self-corrected: when Lex Fridman pressed him—"Can AI replicate a company as complex as NVIDIA?"—Huang answered flatly, "The probability is zero." Even 100,000 AI agents could not rebuild NVIDIA. That self-correction is crucial: it shows Huang holds a clear-eyed view of AI's capabilities[s14].

The timing of the statement is also intriguing. It came during GTC 2026, when NVIDIA announced at least US$1 trillion in Blackwell-plus-Rubin orders. Markets responded positively: NVDA shares rose 1.7 percent, AI-related cryptocurrencies jumped 10 to 20 percent, and on Polymarket the probability of "AGI by 2027" surged from 15 percent to 40 percent. Yet the overwhelming majority of AI researchers believe current systems fall far short of true AGI. Scholars such as Meta's chief AI scientist Yann LeCun have repeatedly accused business leaders of "moving the goalposts"—redefining the target each time AI approaches one[s14][s40].

Huang's layout for future technology can be understood along three dimensions, each pointing to the same core judgment: AI is moving from the digital world into the physical world, and NVIDIA intends to establish an irreplaceable position in both.

First, the generational leap in compute infrastructure—Blackwell and Vera Rubin. The Blackwell architecture (launched March 2024) has 208 billion transistors (a dual-die design), fifth-generation Tensor Cores, native FP4 support and up to 30 times the inference performance of the H100. Vera Rubin (launched March 2026) goes further still: developed by all 40,000 of the company's engineers, it is "the most ambitious undertaking in NVIDIA's history." The Vera Rubin GPU carries 288GB of HBM4 memory; the Vera CPU uses an 88-core Arm architecture with 1.8 times the performance of x86. Huang announced that Vera Rubin has entered full volume production, with the Taiwanese supply-chain footprint twice that of the previous generation. From Blackwell to Vera Rubin, NVIDIA maintains its astonishing rhythm of a new architecture every two years[s7][s11][s39].

Second, physical AI—the next ChatGPT moment. At GTC 2026 Huang declared that "the ChatGPT moment for physical AI is coming," launching the Cosmos 3 physical-AI platform and betting on robotics and AI in the physical world. He believes AI exists not only in the digital world (language models, image generation) but will reach deeply into the physical world—factories, warehouses, cars and robots all need AI to perceive, understand and manipulate physical environments. The Cosmos platform aims to give robots an understanding of the physical world—the next great frontier of AI[s14].

Third, tokenomics—the new economic model of the AI age. Huang has proposed a wholly new economic framework: "Tokens are the new currency of the AI era." In his telling, every AI-generated token—a word, a pixel, a decision—carries economic value; they are the micro-units of "revenue." Data centres have turned from "storage facilities" into "token-production factories." The world is racing to build AI factories in "the largest infrastructure build-out in human history." In that framework NVIDIA is not a chip company but a company supplying the "power-generation equipment" for the AI economy[s14].

On May 13, 2026, NVIDIA's market value broke US$5.5 trillion intraday, exceeding Germany's entire annual GDP—a single company worth more than the full-year economic output of a major economy, something without precedent in the history of business. On August 27, FY2027 Q2 results showed revenue of US$96.2 billion (up 106 percent year on year), and Huang declared that "AI has reached its tipping point." NVIDIA's story is far from over—as he often says: "We are always thirty days away from going out of business."[s11][s19]

## 卷尾 Editorial Conclusion

回望黄仁勋六十余年的人生轨迹，他始终在做同一件事——在所有人都看到风险的地方，押下别人看不懂的赌注。在NV1失败、公司濒临破产时，他飞到日本告诉世嘉要停止项目，然后请求对方仍然全额付款——这不是赌徒的蛮勇，而是基于技术判断的孤注一掷[s9]。在所有人只关心游戏画面的年代，他押注CUDA让GPU可以做通用计算，市值暴跌80%，坚持十年等来AlexNet[s10][s11]。在ChatGPT还没有出现的年代，他亲自把第一台DGX-1超级计算机装进车里，开到旧金山送给OpenAI[s17]。他说过一句著名的话："我们投资于零亿美元的市场。"这不是一句漂亮的口号，而是他一生的行事逻辑——在别人看不见机会的地方，用信念建立基础设施，然后等待世界赶上来[s18]。如今，英伟达市值突破5万亿美元，黄仁勋的个人财富接近两千亿美元，但他依然每天凌晨4点起床，依然穿那件标志性的黑色皮衣，依然说自己"距离倒闭永远只有30天"[s19][s20]。从4万美元的银行贷款到5万亿美元的科技帝国，从Denny's餐厅的卡座到全球科技舞台的中心，从肯塔基的厕所到硅谷的巅峰——黄仁勋的故事，是一个关于信念、忍耐和复利的故事。在剑桥大学的Stephen Hawking Fellowship演讲上，他说"当CEO是一生的牺牲"。在Lex Fridman播客上，他说"我不想失败的动力远大于想成功的动力"[s21][s22]。或许，这个曾经被送到感化学校的男孩，从未忘记那种随时可能失去一切的感觉。正是这种恐惧，驱动着他不断地奔跑——不是为了成功，而是为了不失败。正如他常说的："不要无聊，也不要被解雇。"这大概是一个从底层爬上来的移民，能给出的最朴素也最深刻的人生哲学[s3][s22]。

Looking back over more than six decades of Jensen Huang's life, he has been doing the same thing all along: where everyone else sees only risk, he places bets that others cannot understand. When the NV1 failed and the company neared bankruptcy, he flew to Japan to tell Sega he was halting the project—and then asked Sega to pay in full anyway. That was not a gambler's bravado but an all-in wager grounded in technical judgment[s9]. In an era when everyone cared only about game graphics, he bet on CUDA so that GPUs could do general-purpose computing; the market value plunged 80 percent, and he held on for a decade until AlexNet arrived[s10][s11]. Before ChatGPT existed, he personally loaded the first DGX-1 supercomputer into his car and drove it to San Francisco to give to OpenAI[s17]. He famously said, "We invest in zero-billion-dollar markets." That is no pretty slogan. It is the operating logic of his life: where others see no opportunity, build infrastructure on conviction, and wait for the world to catch up[s18]. Today NVIDIA's market value exceeds US$5 trillion and Huang's personal fortune approaches US$200 billion, yet he still rises at four in the morning, still wears his signature black leather jacket, still says the company is "always thirty days away from going out of business"[s19][s20]. From a US$40,000 bank loan to a US$5-trillion technology empire, from a Denny's booth to the centre of the global technology stage, from Kentucky toilets to the summit of Silicon Valley, Huang's story is one of conviction, endurance and compounding. Delivering the Stephen Hawking Fellowship lecture at Cambridge, he said, "To be a CEO is a lifetime of sacrifice." On the Lex Fridman podcast he said, "My motivation not to fail is far greater than my motivation to succeed"[s21][s22]. Perhaps the boy once sent to a reform school never forgot the feeling that everything could be taken away at any moment. It is that fear that keeps him running—not to succeed, but not to fail. As he often says, "Don't be bored, and don't get fired." It may be the plainest, and the most profound, philosophy of life that an immigrant who climbed from the bottom can offer[s3][s22].

## 金句 Pull Quote

> 伟大来自品格。而品格不是聪明人天生的，是苦过的人磨出来的。我希望你们经历足够多的痛苦和磨难。

> "Greatness comes from character. And character is not something smart people are born with—it is forged by those who have suffered. I hope you experience enough pain and hardship." — Jensen Huang, Stanford University speech, 2024

## 履历时间线 Career Timeline

- **1963-02-17** 出生于中国台湾台北市 / Born in Taipei, Taiwan
  - 1963年2月17日，黄仁勋出生于中国台湾台北市，后随家人迁居台南。父亲黄兴泰是化学工程师，母亲罗采秀是教师。母亲每天从字典中选10个单词教儿子英语，她自己不懂英文。4岁半时逛夜市被利刃划伤脸部留疤[s1][s2]。
  - EN: On February 17, 1963, Jensen Huang was born in Taipei, Taiwan, and later moved with his family to Tainan. His father, Huang Hsing-tai, was a chemical engineer and his mother, Lo Tsai-hsiu, was a teacher. Speaking no English herself, his mother picked ten words from the dictionary every day to teach her sons English. At four and a half he was cut in the face by a blade at a night market, leaving a scar[s1][s2].
- **1968** 移居泰国 / Moves to Thailand
  - 5岁时因父亲工作调动全家迁往泰国曼谷，就读Ruamrudee国际学校。在泰国约四年。曾与哥哥将打火机燃料倒入游泳池点燃[s3][s4]。
  - EN: At five, after his father was transferred for work, the family moved to Bangkok, Thailand, where he attended Ruamrudee International School. He spent about four years in Thailand. Once he and his older brother poured lighter fluid into a swimming pool and set it alight[s3][s4].
- **1973** 被送往美国肯塔基 / Sent to Kentucky, USA
  - 9-10岁的黄仁勋和哥哥被送到华盛顿州塔科马市舅舅家。舅舅误将他们送入肯塔基州Oneida Baptist Institute——一所面向问题青少年的宗教寄宿学校。他是该校历史上年龄最小的寄宿生，每天打扫厕所，遭受种族欺凌[s3][s4]。
  - EN: Nine- or ten-year-old Huang and his brother were sent to live with an uncle in Tacoma, Washington. The uncle mistakenly enrolled them at Kentucky's Oneida Baptist Institute, a religious boarding school for troubled youth. He was the youngest boarder in the school's history, cleaned toilets every day and endured racial bullying[s3][s4].
- **1975** 全家团聚俄勒冈 / Family reunited in Oregon
  - 到肯塔基约两年后父母移民俄勒冈州比弗顿市，全家团聚。进入阿罗哈高中，连跳两级。成为全美排名的乒乓球选手，14岁登上《体育画报》。加入游泳队，学会打乒乓球[s1][s4]。
  - EN: About two years after arriving in Kentucky, his parents emigrated to Beaverton, Oregon, and the family was reunited. He entered Aloha High School and skipped two grades. He became a nationally ranked table-tennis player and appeared in Sports Illustrated at fourteen. He joined the swimming team and learned to play table tennis[s1][s4].
- **1977** 首次接触计算机 / First encounter with computers
  - 学校购买Apple II电脑，黄仁勋用BASIC编写了自己的贪吃蛇游戏版本，还玩Super Star Trek文字游戏——这是他与计算机世界的第一次亲密接触[s1]。
  - EN: His school bought an Apple II computer, and Huang wrote his own version of the Snake game in BASIC and played the text-based Super Star Trek game—his first close encounter with the world of computers[s1].
- **1978** 开始在Denny's打工 / Starts working at Denny's
  - 从15岁起在当地Denny's餐厅做夜班——洗碗工、勤杂工、服务员。他后来说自己"洗过很多很多厕所"。这段经历塑造了从最底层做起的职业观[s5][s6]。
  - EN: From the age of fifteen he worked the graveyard shift at a local Denny's restaurant—as dishwasher, busboy and waiter. He later said he had "washed a lot, a lot of toilets." The experience shaped his outlook of starting from the very bottom[s5][s6].
- **1979** 16岁考入大学 / Enters university at 16
  - 16岁考入俄勒冈州立大学电子工程专业，选择原因是"州内学费便宜"。在实验室结识Lori Mills（实验课搭档），向她承诺"30岁时成为一家公司的CEO"[s15][s16]。
  - EN: At sixteen he was admitted to Oregon State University to study electrical engineering, choosing it because "in-state tuition was cheap." In a laboratory he met Lori Mills (his lab partner) and promised her he would "become the CEO of a company by the time I was thirty"[s15][s16].
- **1984** 毕业加入AMD / Graduates and joins AMD
  - 大学毕业获电子工程学士学位（BSEE），加入AMD担任芯片设计师。开始学普通话以便与中国光罩工人沟通[s1][s5]。
  - EN: He graduated with a bachelor's degree in electrical engineering (BSEE) and joined AMD as a chip designer. He began learning Mandarin so that he could communicate with the company's Chinese mask workers[s1][s5].
- **1985** 加入LSI Logic / Joins LSI Logic
  - 转职至LSI Logic公司，结识Chris Malachowsky和Curtis Priem。在销售部门工作过，认为这是"最佳职业选择之一"。开始利用工作之余攻读斯坦福硕士[s5][s7]。
  - EN: He moved to LSI Logic, where he met Chris Malachowsky and Curtis Priem. He worked for a time in the sales department, which he considered "one of the best career choices I ever made." He began studying part-time for a Stanford master's degree[s5][s7].
- **1990** 获斯坦福硕士学位 / Earns Stanford master's degree
  - 获得斯坦福大学电子工程硕士学位（MSEE），系统接触图形渲染技术，为日后创办英伟达奠定技术基础[s1][s7]。
  - EN: He received a master's degree in electrical engineering (MSEE) from Stanford University, where he was systematically exposed to graphics-rendering technology, laying the technical foundation for founding NVIDIA later[s1][s7].
- **1993-04-05** 创立英伟达 / Founds NVIDIA
  - 1993年4月5日，与Chris Malachowsky、Curtis Priem在加州圣何塞一家Denny's餐厅共同创立NVIDIA。约4万美元银行贷款启动，公司注册时律师拿走黄仁勋口袋里仅有的200美元现金，又向两位合伙人各要200美元——名义注册资本600美元。后续从红杉资本和Sutter Hill获200万美元种子轮融资，投后估值600万。公司名源自拉丁语"invidia"（嫉妒）。30岁的黄仁勋任CEO，兑现对妻子的承诺[s7][s8]。
  - EN: On April 5, 1993, he co-founded NVIDIA with Chris Malachowsky and Curtis Priem at a Denny's restaurant in San Jose, California. It started with about US$40,000 in bank loans; when the company was incorporated the lawyer took the only US$200 cash in Huang's pocket, and he asked each partner for another US$200—notional registered capital of US$600. A later US$2 million seed round from Sequoia Capital and Sutter Hill Ventures valued the company at US$6 million post-money. The name comes from the Latin "invidia" (envy). At thirty, Huang became CEO, keeping his promise to his wife[s7][s8].
- **1995** NV1失败，濒临破产 / NV1 failure, near bankruptcy
  - 首款芯片NV1采用四边形渲染技术路线，不兼容微软Direct3D标准，市场反应惨淡。公司从约100人裁至约30-35人（裁员70%），银行账户不足100万美元。黄仁勋抵押房产为员工发工资——"我们距离倒闭永远只有30天"这句话的源头就在这里。这是英伟达历史上最接近彻底失败的时刻[s9]。
  - EN: The first chip, the NV1, used a quadrilateral-rendering approach and was incompatible with Microsoft's Direct3D standard, drawing a dismal market response. The company cut staff from about 100 to roughly 30-35 (a 70 percent layoff), with less than US$1 million in the bank. Huang mortgaged his house to make payroll—the origin of the line "we are always thirty days away from going out of business." It was the closest NVIDIA ever came to total failure[s9].
- **1996** 世嘉谈判：起死回生 / The Sega negotiation: back from the dead
  - 黄仁勋飞赴日本面见世嘉CEO，主动告知停止开发Dreamcast芯片，请求对方仍付全额合同款（500-700万美元）——即使世嘉不会收到产品。这笔钱拯救了英伟达。黄仁勋说"如果他说不，我们就会倒闭"。他从这次经历学到：当你没有筹码时，信念就是唯一的筹码[s10]。
  - EN: Huang flew to Japan to meet Sega's CEO, told him unprompted that NVIDIA was halting development of the Dreamcast chip, and asked Sega to pay the full contract price anyway (US$5-7 million)—even though Sega would never receive the product. The money saved NVIDIA. Huang said, "If he had said no, we would have gone out of business." The lesson he drew: when you have no chips to play, conviction is your only chip[s10].
- **1997-04** RIVA 128逆转 / The RIVA 128 turnaround
  - 推出RIVA 128（NV3），全球首款128位3D处理器，兼容Direct3D。四个月售出百万块，公司首度盈利。营收从1996年350万美元跃升至5,750万美元[s5][s9]。
  - EN: NVIDIA launched the RIVA 128 (NV3), the world's first 128-bit 3D processor, compatible with Direct3D. It sold a million units in four months and the company turned profitable for the first time. Revenue jumped from US$3.5 million in 1996 to US$57.5 million[s5][s9].
- **1999-01-22** 纳斯达克上市 / NASDAQ IPO
  - 英伟达在纳斯达克上市（NVDA），发行价$12/股，IPO估值约6亿美元。同年推出GeForce 256，首次提出"GPU"概念，定义全新计算品类[s9][s11]。
  - EN: NVIDIA listed on NASDAQ (NVDA) at US$12 a share, with an IPO valuation of about US$600 million. The same year it launched the GeForce 256 and introduced the concept of the "GPU," defining an entirely new computing category[s9][s11].
- **2000** Xbox合作与收购3dfx / Xbox deal and 3dfx acquisition
  - 2000年是英伟达走出显卡价格战、奠定行业霸主地位的一年。3月，公司击败众多对手，被微软选定为新一代游戏主机Xbox的图形处理器供应商，微软预付2亿美元，总合同金额约5亿美元——相当于当时英伟达全年销售额的规模；9月又拿下Xbox媒体通信处理器（MCP）订单[s7][s11]。4月起公司相继推出GeForce2 GTS（全球首款逐像素着色GPU）、GeForce2 MX、首款移动GPU GeForce2 Go以及Quadro2工作站系列，产品梯队全面铺开[s7]。12月15日，英伟达与昔日显卡霸主3dfx签署最终协议，以7000万美元现金加100万股股票（合计约1.12亿美元）收购其全部图形核心资产，包括专利、"3dfx"与"Voodoo"商标及约百名工程师，双方专利诉讼同时和解撤销，图形芯片行业就此进入英伟达主导的时代[s7][s11]。
  - EN: NVIDIA listed on NASDAQ (NVDA) at US$12 a share, issuing 3.5 million shares at an IPO valuation of about US$600 million. The same year it launched the GeForce 256, introduced the "GPU" concept for the first time, integrated T&L hardware acceleration and defined an entirely new computing category. From US$40,000 in startup capital to a US$600 million IPO valuation took less than six years[s9][s11].
- **2006** CUDA发布 / CUDA launched
  - 发布CUDA并行计算平台，允许用C语言直接调用GPU进行通用计算。决定在每块GeForce显卡上植入CUDA，生产成本增加约50%，市值暴跌约80%。开始"沉默的十年"[s10][s12]。
  - EN: NVIDIA launched the CUDA parallel-computing platform, allowing developers to use the C language to call on GPUs directly for general-purpose computing. The decision to build CUDA into every GeForce card raised production costs by about 50 percent and crashed the market value by roughly 80 percent. The "silent decade" began[s10][s12].
- **2012-09** AlexNet：AI大爆炸 / AlexNet: the AI Big Bang
  - 多伦多大学Alex Krizhevsky用两块GTX 580显卡（每块约500美元）训练的AlexNet在ImageNet中取得15.3%错误率（第二名26.2%），断层式突破。黄仁勋称之为"AI的大爆炸"。所有深度学习实验室开始购买NVIDIA GPU，CUDA十年投入终获回报——从几百个开发者到整个AI行业[s12][s13]。
  - EN: AlexNet, trained by the University of Toronto's Alex Krizhevsky on two GTX 580 cards (about US$500 each), scored a 15.3 percent error rate in ImageNet (versus 26.2 percent for the runner-up), a landslide breakthrough. Huang called it "the Big Bang of AI." Every deep-learning lab began buying NVIDIA GPUs, and a decade of CUDA investment finally paid off—from a few hundred developers to an entire AI industry[s12][s13].
- **2016** Pascal架构与DGX-1 / Pascal architecture and DGX-1
  - 推出Pascal架构（P100），首款专为深度学习设计的GPU。推出DGX-1超级计算机，黄仁勋亲自把第一台开车送到OpenAI。数据中心业务开始快速增长[s17]。
  - EN: NVIDIA launched the Pascal architecture (P100), the first GPU designed specifically for deep learning, and unveiled the DGX-1 supercomputer; Huang personally drove the first unit to OpenAI. The data-centre business began to grow rapidly[s17].
- **2022-11** ChatGPT引爆AI算力需求 / ChatGPT ignites demand for AI compute
  - ChatGPT发布，H100 GPU供不应求。同月放弃400亿美元收购Arm交易。数据中心收入开始指数级增长[s11]。
  - EN: ChatGPT was released and H100 GPUs were in desperate short supply. In the same month the company abandoned the US$40 billion acquisition of Arm. Data-centre revenue began its exponential climb[s11].
- **2023-05** 市值突破1万亿美元 / Market value breaks US$1 trillion
  - 英伟达市值首次突破1万亿美元，成为首家破万亿的芯片公司。全年市值增长240%，年底达1.22万亿美元。ChatGPT引发的AI算力需求是核心驱动力——从FY2023约150亿美元到FY2024的475亿美元数据中心收入[s11]。
  - EN: NVIDIA's market capitalisation crossed US$1 trillion for the first time, making it the first chip company to reach the milestone. Its value rose 240 percent over the year, reaching US$1.22 trillion by year-end. The AI-compute demand unleashed by ChatGPT was the core driver—data-centre revenue going from about US$15 billion in FY2023 to US$47.5 billion in FY2024[s11].
- **2024-03** Blackwell架构发布 / Blackwell architecture launched
  - GTC发布Blackwell架构（B200），2,080亿晶体管（双die设计），第五代Tensor Core，推理性能较H100提升最高30倍。6月市值突破3万亿，一度超越微软成为全球市值最高公司。黄仁勋第二次入选《时代》100人，当选美国工程院院士[s11]。
  - EN: GTC saw the launch of the Blackwell architecture (B200), with 208 billion transistors (a dual-die design), fifth-generation Tensor Cores and up to 30 times the inference performance of the H100. In June the market value broke US$3 trillion, briefly overtaking Microsoft as the world's most valuable company. Huang was named to Time's 100 list for the second time and elected to the US National Academy of Engineering[s11].
- **2025-10-29** 市值突破5万亿美元 / Market value breaks US$5 trillion
  - 成为历史上首家收盘市值突破5万亿美元的公司。入选《时代》2025年度人物（"AI的缔造者"），位列《财富》全球100位最具影响力商界人士第一位。11月剑桥大学获Stephen Hawking Fellowship，说"当CEO是一生的牺牲"。12月特朗普批准H200对华出口（25%分成条件）[s11][s21]。
  - EN: NVIDIA became the first company in history to close with a market value above US$5 trillion. Huang was named Time's 2025 Person of the Year ("the builder of AI") and ranked first on Fortune's list of the world's 100 most influential business leaders. In November he received the Stephen Hawking Fellowship at Cambridge, saying "to be a CEO is a lifetime of sacrifice." In December the Trump administration approved H200 exports to China (with a 25 percent revenue-share condition)[s11][s21].
- **2026-01** CES发布RTX 50系列 / RTX 50 series unveiled at CES
  - CES发表主题演讲，发布GeForce RTX 50系列、Cosmos基础模型。宣布Vera Rubin所有芯片已流片回归，硅验证完成[s11]。
  - EN: He delivered the CES keynote, launching the GeForce RTX 50 series and the Cosmos foundation models. He announced that all Vera Rubin chips had returned from tape-out and that silicon validation was complete[s11].
- **2026-03** AGI宣言与Vera Rubin发布 / AGI declaration and Vera Rubin launch
  - GTC 2026发布Vera Rubin架构，动用40,000名工程师。在Lex Fridman播客中宣称"我认为我们已经实现了AGI"，引发全球讨论[s11][s14]。
  - EN: GTC 2026 saw the launch of the Vera Rubin architecture, mobilising 40,000 engineers. On the Lex Fridman podcast he declared, "I think we've achieved AGI," sparking global debate[s11][s14].
- **2026-05-13** 市值盘中破5.5万亿 / Market value tops US$5.5 trillion intraday
  - 市值盘中突破5.5万亿美元，超越德国全年GDP。FY2027 Q2营收962亿美元（同比+106%），宣告"AI已到达转折点"[s11]。
  - EN: Market value broke US$5.5 trillion intraday, exceeding Germany's entire annual GDP. FY2027 Q2 revenue came to US$96.2 billion (up 106 percent year on year), and he declared that "AI has reached its tipping point"[s11].

## 常问问答 FAQ

**Q1: 黄仁勋是哪里人？他的国籍是什么？**

A: 黄仁勋出生于中国台湾台北市，在台南市长大，祖籍浙江省青田县。他1973年（9-10岁）移居美国，现为美籍华人。父亲黄兴泰是化学工程师，母亲罗采秀是教师，家中说闽南语（台语）。黄仁勋与AMD CEO苏姿丰（Lisa Su）是表亲关系——黄母是苏外祖父的幼妹，即first cousins once removed关系。两人分别执掌着全球领先的两家GPU公司，既是亲戚又是竞争对手，为科技行业增添了一抹人情味[s1][s2]。

**Q1 (EN): Where is Jensen Huang from, and what is his nationality?**

A: Jensen Huang was born in Taipei, Taiwan, and grew up in Tainan; his family's ancestral home is Qingtian County, Zhejiang Province. He moved to the United States in 1973 at the age of nine or ten and is now a Taiwanese-born American citizen. His father, Huang Hsing-tai, was a chemical engineer and his mother, Lo Tsai-hsiu, was a teacher; the family spoke Hokkien (Taiwanese) at home. Huang is also related to AMD CEO Lisa Su: his mother was the youngest sister of Su's maternal grandfather, making the two first cousins once removed. They respectively run the two largest GPU companies in the world—relatives and rivals at once, adding a touch of human warmth to the technology industry[s1][s2].

**Q2: 黄仁勋今年多大？他的教育背景如何？**

A: 黄仁勋1963年2月17日出生，截至2026年9月为63岁。教育经历：16岁考入俄勒冈州立大学电子工程专业（连跳两级，16岁毕业），1984年获学士学位（BSEE）。工作期间利用业余时间攻读斯坦福大学硕士，1990年获电子工程硕士学位（MSEE）。此外拥有中国台湾阳明交通大学、中国台湾大学、俄勒冈州立大学、华中科技大学、瑞典林雪平大学等五所高校的荣誉博士学位[s1][s7]。

**Q2 (EN): How old is Jensen Huang, and what is his educational background?**

A: Jensen Huang was born on February 17, 1963, making him 63 as of September 2026. His education: he was admitted to Oregon State University at 16 to study electrical engineering (having skipped two grades and graduating high school at 16), and he earned his bachelor's degree (BSEE) in 1984. While working full time he studied part-time for a master's at Stanford University, receiving an MSEE in electrical engineering in 1990. He also holds honorary doctorates from five universities: National Yang Ming Chiao Tung University in Taiwan, National Taiwan University, Oregon State University, Huazhong University of Science and Technology in China and Linkoping University in Sweden[s1][s7].

**Q3: 黄仁勋的妻子是谁？有孩子吗？**

A: 黄仁勋的妻子叫Lori Mills，两人在俄勒冈州立大学相识——她是黄仁勋工程实验课的搭档（lab partner），当时电气工程系250名学生中只有3位女生。两人1985年结婚，至今携手四十年。育有一子一女：长子黄盛斌（Spencer，产品经理），小女儿黄敏珊（Madison，产品营销总监），均在英伟达工作。Spencer曾在中国台湾台北开设酒吧入选亚洲Top 50，2021年关闭后加入英伟达，颇有乃父Denny's打工的创业精神[s1][s2]。

**Q3 (EN): Who is Jensen Huang's wife, and does he have children?**

A: Jensen Huang's wife is Lori Mills, whom he met at Oregon State University—she was his engineering lab partner, at a time when only three of the electrical-engineering department's 250 students were women. They married in 1985 and have now been together for forty years. They have a son and a daughter: their elder son, Spencer Huang, is a product manager, and their younger daughter, Madison Huang, is a director of product marketing; both work at NVIDIA. Spencer once ran a bar in Taipei, Taiwan, that made Asia's Top 50 Bars list; after it closed in 2021 he joined NVIDIA—very much in the entrepreneurial spirit of his father's Denny's days[s1][s2].

**Q4: 黄仁勋有多少钱？英伟达市值多少？**

A: 黄仁勋持有约3-3.5%的英伟达股份，财富几乎全部与股价挂钩。净资产随股价大幅波动：2025年11月Bloomberg Billionaires Index约1,650亿美元，2026年8月约1,968亿美元（Forbes Argentina数据），位列全球富豪榜第六位左右。英伟达市值方面：2026年5月13日盘中突破5.5万亿美元（超越德国全年GDP），2026年7月约5.15万亿美元。从1万亿到5万亿仅用两年半，创商业史增长奇迹[s26]。

**Q4 (EN): How much money does Jensen Huang have, and what is NVIDIA's market value?**

A: Jensen Huang holds roughly 3 to 3.5 percent of NVIDIA, so his fortune is almost entirely tied to the share price. His net worth swings with the stock: around US$165 billion on the Bloomberg Billionaires Index in November 2025 and about US$196.8 billion in August 2026 (per Forbes Argentina data), placing him around sixth on the global rich list. As for NVIDIA's market capitalisation: it broke US$5.5 trillion intraday on May 13, 2026 (exceeding Germany's entire annual GDP) and stood at roughly US$5.15 trillion in July 2026. The climb from US$1 trillion to US$5 trillion took only two and a half years, a growth miracle in business history[s26].

**Q5: 为什么黄仁勋总穿黑色皮衣？**

A: 自2013年前后起，黄仁勋以黑色皮衣、黑衬衫搭配牛仔裤的造型出现在公众场合，被中国网友称为"皮衣黄"。这已成为他个人品牌的核心符号，与乔布斯的黑高领、扎克伯格的灰T恤齐名。皮衣品牌从未公开披露（Tom Ford说法为推测，未获证实）。在韩国"Jensen Huang jacket"搜索量激增130%。2024年中国台湾媒体创造"Jensanity"一词形容围绕他的狂热现象。Mark Zuckerberg称他是"科技界的Taylor Swift"[s16]。

**Q5 (EN): Why does Jensen Huang always wear a black leather jacket?**

A: From around 2013 onward, Jensen Huang began appearing in public in a black leather jacket, black shirt and jeans, earning him the Chinese nickname "Leather Jacket Huang." It has become the core symbol of his personal brand, ranked alongside Steve Jobs's black turtleneck and Mark Zuckerberg's grey T-shirt. The jacket's brand has never been publicly disclosed (the Tom Ford claim is speculation and unverified). During a visit to South Korea, searches for "Jensen Huang jacket" surged 130 percent. In 2024 Taiwanese media coined the word "Jensanity" to describe the fan frenzy around him, and Mark Zuckerberg called him "the Taylor Swift of the tech world"[s16].

**Q6: CUDA是什么？为什么黄仁勋要押注CUDA？**

A: CUDA是英伟达2006年推出的并行计算平台，允许开发者用C语言等通用编程语言直接调用GPU进行通用计算。黄仁勋坚信"未来计算是并行的"，决定在每块GeForce显卡上都植入CUDA功能，包括最便宜的游戏卡。代价巨大：生产成本增加约50%，市值暴跌约80%（从80-120亿跌至15-20亿美元）。华尔街质疑"疯了"。但黄仁勋坚持十年——亲自去大学教课、赠送开发卡、每年投入数亿美元。2012年AlexNet证明了GPU深度学习价值，CUDA成为AI基石。如今500万+开发者，300+库，3700+应用。竞争对手可复制芯片，但无法复制十年生态积累[s10][s12]。

**Q6 (EN): What is CUDA, and why did Jensen Huang bet on it?**

A: CUDA is NVIDIA's parallel-computing platform, launched in 2006, which lets developers use general-purpose languages such as C to call on GPUs directly for general-purpose computing. Convinced that "the future of computing is parallel," Huang decided to build CUDA into every GeForce card, including the cheapest gaming models. The cost was enormous: production costs rose about 50 percent and the market value plunged roughly 80 percent (from US$8-12 billion to US$1.5-2 billion). Wall Street asked if he had gone mad. But Huang held on for a decade—teaching at universities in person, giving away development cards and spending hundreds of millions of dollars a year. In 2012 AlexNet proved the value of GPUs for deep learning, and CUDA became the cornerstone of AI. Today it has more than 5 million developers, over 300 libraries and more than 3,700 applications. Rivals can copy the chips, but they cannot copy a decade of ecosystem accumulation[s10][s12].

**Q7: 黄仁勋的管理风格是怎样的？**

A: 黄仁勋的管理风格极为独特：直接管理50-60个下属（标准CEO只有8-12个），不做一对一会议，推行全员T5T五点邮件制度，公开批评而非私下批评，"痛苦容忍度"哲学。每天工作12-14小时，每周7天，凌晨4点起床。被传记作者描述为"被恐惧和内疚驱动"——"不想失败的动力远大于想成功的动力"。自称"firing-averse"但喜欢"torture people into greatness"。尽管严苛，流失率仅约2.5%（FY2025），远低于硅谷平均水平[s20][s22][s23][s25]。

**Q7 (EN): What is Jensen Huang's management style?**

A: Jensen Huang's management style is highly distinctive: he directly manages 50 to 60 subordinates (a standard CEO has 8 to 12), holds no one-on-one meetings, runs the company-wide T5T (Top 5 Things) email system, criticises in public rather than in private, and preaches a philosophy of high "pain tolerance." He works 12 to 14 hours a day, seven days a week, and rises at four in the morning. Biographers describe him as "driven by fear and guilt"—"my motivation not to fail is far greater than my motivation to succeed." He calls himself "firing-averse" but fond of "torturing people into greatness." For all the rigour, attrition was only about 2.5 percent in FY2025, far below the Silicon Valley average[s20][s22][s23][s25].

**Q8: 黄仁勋为什么说AGI已经实现了？**

A: 2026年3月，黄仁勋在Lex Fridman播客第494期中说"我认为我们已经实现了AGI"。关键是他的AGI定义是"经济性的"而非学术性的——标准是"一个能自主创造价值10亿美元服务的AI"，而非"在所有认知任务上达到人类水平"。他同时承认10万个AI智能体也无法重建英伟达（"概率为零"）。学术界普遍认为当前AI远未达到真正AGI。Yann LeCun等学者质疑商业领袖在"移动球门柱"。时机耐人寻味——正值GTC大会宣布万亿美元订单期间[s14][s40]。

**Q8 (EN): Why does Jensen Huang say that AGI has already been achieved?**

A: In March 2026, on episode 494 of the Lex Fridman podcast, Jensen Huang said, "I think we've achieved AGI." The key is that his definition of AGI is "economic" rather than academic—his benchmark is "an AI that can autonomously create a service worth US$1 billion," rather than one that "reaches human level on all cognitive tasks." He also conceded that even 100,000 AI agents could not rebuild NVIDIA ("the probability is zero"). The academic mainstream believes today's AI falls far short of true AGI, and scholars such as Yann LeCun accuse business leaders of "moving the goalposts." The timing was striking: the remark came during GTC, alongside the announcement of a trillion dollars in orders[s14][s40].

**Q9: 英伟达在中国市场的现状如何？**

A: 受美国出口管制影响，英伟达在中国高端AI芯片市场份额从95%骤降至接近零（2025年底数据）。H20特供芯片（算力约H100的1/6）曾是唯一合规选择，2025年4月许可证也被叫停，7月恢复。12月H200获批对华出口（附25%收入分成条件），2026年1月改为逐案审查。有少量H200交付但总量受限。中国国产替代加速——华为昇腾已达H200级性能，预计2026年占近90%中国高端AI芯片市场。黄仁勋一直反对管制[s27][s28]。

**Q9 (EN): What is NVIDIA's current position in the China market?**

A: Under US export controls, NVIDIA's share of China's high-end AI-chip market has plunged from 95 percent to nearly zero (data as of end-2025). The China-specific H20 chip (about one-sixth the compute of the H100) was once the only compliant option; its export licence was halted in April 2025 and restored in July. In December the H200 was approved for export to China (with a 25 percent revenue-share condition), and in January 2026 approvals switched to a case-by-case basis. Small quantities of H200s have been delivered but volumes remain severely limited. China's domestic substitution is accelerating—Huawei's Ascend has reached H200-class performance and is expected to take nearly 90 percent of the Chinese high-end AI-chip market in 2026. Huang has consistently opposed the controls[s27][s28].

**Q10: 黄仁勋获得过哪些重要荣誉？**

A: 主要荣誉：IEEE Robert N. Noyce奖（半导体最高荣誉）、IEEE创始人勋章、IEEE荣誉勋章（2026年）、《财富》全球最佳CEO、《经济学人》年度最佳CEO、Brand Finance全球100位最佳CEO第三名（2025年）、《时代》周刊100位最具影响力人物（2021年、2024年两次入选）、2025年伊丽莎白女王工程奖、2024年美国工程院院士、2025年《时代》年度人物（"AI的缔造者"）、2025年《财富》全球100位最具影响力商界人士第一位、2026年PCAST成员[s1]。

**Q10 (EN): What major honours has Jensen Huang received?**

A: His principal honours include: the IEEE Robert N. Noyce Medal (the semiconductor industry's highest honour), the IEEE Founders Medal and the IEEE Medal of Honor (2026); Fortune's Global Best CEO; The Economist's CEO of the Year; third place on Brand Finance's list of the world's top 100 CEOs (2025); inclusion in Time's list of the 100 most influential people (twice, in 2021 and 2024); the 2025 Queen Elizabeth Prize for Engineering; election to the US National Academy of Engineering in 2024; Time's 2025 Person of the Year (as "the builder of AI"); first place on Fortune's 2025 list of the world's 100 most influential business leaders; and membership of PCAST in 2026[s1].

**Q11: 英伟达为什么叫NVIDIA？**

A: 公司名源自拉丁语"invidia"，意为"嫉妒"（罗马神话中嫉妒女神Invidia）。发音为en-VID-ee-ah。三位创始人考虑过PixelPushers（太直白）、Rendition（已被使用）等名字，最终选择NVIDIA。绿色眼睛Logo，螺旋图案代表无限计算，暗含"要让同行仰望、嫉妒"的野心。早期文件拼写为"nVidia"（小写n），1997年后统一为"NVIDIA"[s1][s8]。

**Q11 (EN): Why is the company called NVIDIA?**

A: The company name derives from the Latin word "invidia," meaning envy (Invidia being the Roman goddess of envy); it is pronounced en-VID-ee-ah. The three founders considered names such as PixelPushers (too blunt) and Rendition (already taken) before settling on NVIDIA. The green-eyed logo, with its spiral representing infinite computation, carries the ambition of making rivals look up in envy. Early documents spelled the name "nVidia" with a lowercase n; after 1997 it was standardised as the all-caps "NVIDIA"[s1][s8].

**Q12: 黄仁勋在肯塔基的少年经历是怎样的？**

A: 1973年，9-10岁的黄仁勋和哥哥被送到美国。舅舅误将他们送入肯塔基州Oneida Baptist Institute——一所面向问题青少年的宗教寄宿学校（非"精英学校"）。他是该校历史上年龄最小的寄宿生。每天打扫厕所，室友是17岁问题少年，遭受种族欺凌。但他加入游泳队、学会乒乓球，14岁登上《体育画报》。他说在肯塔基的记忆"比其他任何记忆都更清晰"。2019年向该校捐赠200万美元修建Jen-Hsun Huang Hall[s3][s4]。

**Q12 (EN): What was Jensen Huang's boyhood experience in Kentucky like?**

A: In 1973, nine- or ten-year-old Jensen Huang and his brother were sent to the United States. Their uncle mistakenly enrolled them at the Oneida Baptist Institute in Kentucky—a religious boarding school for troubled youth, not an "elite academy." Jensen was the youngest boarder in the school's history. He cleaned toilets every day, roomed with a 17-year-old delinquent and endured racial bullying. Yet he joined the swimming team, learned table tennis and appeared in Sports Illustrated at fourteen. He says his memories of Kentucky are "more vivid than any other memory." In 2019 he donated US$2 million to the school to build Jen-Hsun Huang Hall[s3][s4].

## 来源清单 Sources

1. [s1] NVIDIA官方 · Jensen Huang - Executive Biography — https://nvidianews.nvidia.com/bios/jensen-huang
2. [s2] owiki.org · Jensen Huang — https://www.owiki.org/wiki/Jensen_Huang
3. [s3] tjhammons.com · Jensen Huang: The Genius Story — http://www.tjhammons.com/jensenhuanggenius.html
4. [s4] tryalma.com · Jensen Huang and Their Immigration Story — https://www.tryalma.com/learn/jensen-huang-immigration-story
5. [s5] 新浪财经 · 跨越60年长跑 黄仁勋如何将英伟达锻造为AI时代传奇 — https://finance.sina.com.cn/roll/2026-06-08/doc-iniasfps8828385.shtml
6. [s6] Fortune · Nvidia CEO Jensen Huang says this career path will thrive in the AI era — https://dc.fortune.com/2026/04/29/nvidia-ceo-jensen-huang-engineering-path-to-success-ai-era-gen-z-advice-ieee-medal-of-honor-winner/
7. [s7] NVIDIA官方 · NVIDIA Corporate Timeline — https://www.nvidia.com/en-us/about-nvidia/corporate-timeline/
8. [s8] longtermwiki.com · NVIDIA Facts Database — https://www.longtermwiki.com/organizations/nvidia/facts
9. [s9] dfarq.homeip.net · NVIDIA's IPO on January 22, 1999 — https://dfarq.homeip.net/nvidias-ipo-on-january-22-1999/
10. [s10] The Closer · The Nvidia Doctrine — https://www.thecloser.fm/the-nvidia-doctrine/
11. [s11] AIWiki / NeuralWired / pritamroy.com · NVIDIA: Full Story from $40K Bet to $5 Trillion Empire — https://www.aiwiki.ai/wiki/nvidia
12. [s12] Tech Reader Magazine · How Nvidia's Software Ecosystem (CUDA) Captured the AI Market — https://www.techreadermagazine.com/2026/06/how-nvidias-software-ecosystem-cuda.html
13. [s13] NVIDIA Blog · Sutskever and OpenAI at GTC — https://blogs.nvidia.com/blog/2023/03/22/sutskever-openai-gtc/
14. [s14] The Street / Anthem Creation / podchemy · Nvidia CEO Jensen Huang says we have achieved AGI — https://www.thestreet.com/technology/nvidia-ceo-jensen-huang-says-we-have-achieved-agi
15. [s15] 环球人物 · 从扫厕所到打造全球最贵公司 — https://www.globalpeople.com.cn/n4/2025/0718/c305916-21628647.html
16. [s16] Korea Herald / TVBS · Jensen Huang's leather jacket emerges as fashion icon — https://www.koreaherald.com/business/2025/05/31/jensen-huangs-leather-jacket-emerges-as-fashion-icon/
17. [s17] otontechnology.com · NVIDIA Rise: World's Most Valuable Company — https://otontechnology.com/nvidia-rise-worlds-most-valuable-company/
18. [s18] Stratrix · The Decade of Looking Wrong: NVIDIA CUDA Bet — https://www.stratrix.com/decision-forks/nvidia-cuda-bet
19. [s19] 澎湃新闻 · '偏执狂'黄仁勋，百万富翁制造机 — https://www.thepaper.cn/newsDetail_forward_29956163
20. [s20] Fortune · 60 direct reports, but no 1-on-1 meetings: How Jensen Huang leads Nvidia — https://www.fortune.com/2024/11/12/jensen-huang-nvidia-ceo-leadership-mpp/
21. [s21] Fortune · Nvidia's Jensen Huang says 'to be a CEO is a lifetime of sacrifice' — https://fortune.com/2025/11/17/nvidia-ceo-jensen-huang-sacrifice-leadership-reality/
22. [s22] CNBC · Biographer Stephen Witt on what he learned about Nvidia CEO — https://www.cnbc.com/2025/06/04/jensen-huang-biographer-stephen-witt-what-i-learned-about-nvidia-ceo.html
23. [s23] Insider (Business Insider) · Nvidia is the original hardcore tech company — https://www.insider.com/nvidia-hardcore-intensity-culture-mission-driven-jensen-huang-2025-4
24. [s24] fathomlessgaming.com · An Existential Threat: How the CUDA Bet Almost Destroyed NVIDIA — https://fathomlessgaming.com/an-existential-threat-nvidia-ceo-explains-how-the-cuda-bet-almost-destroyed-the-company-and-why-he-stuck-with-it-anyway/
25. [s25] podchemy.com · Lex Fridman Podcast #494 - Jensen Huang — https://www.podchemy.com/notes/494-jensen-huang-nvidia-the-4-trillion-company-the-ai-revolution-52326234903
26. [s26] Forbes Argentina / Bloomberg Billionaires Index · Jensen Huang's net worth evolution — https://www.forbesargentina.com/millonarios/jensen-huang-supero-zuckerberg-ellison-su-fortuna-ya-roza-us-200000-millones-su-evolucion-2019-n96567
27. [s27] 新浪财经 / 环球人物 / 证券时报 · 英伟达中国市场份额从95%降至接近零 — https://finance.sina.com.cn/jjxw/2025-10-17/doc-infuetkm1614513.shtml
28. [s28] 证券时报 / informedclearly.com · 英伟达重启H200对华生产 — https://stcn.com/article/detail/3683804.html
29. [s29] CNBC / juniorstocks.com · Jensen Huang Sounds the Alarm on the Global AI Race — https://www.juniorstocks.com/we-re-not-far-ahead-of-china-overall-jensen-huang-sounds-the-alarm-on-the-global-ai-race
30. [s30] ACM Communications · NVIDIA at the Center of the Generative AI Ecosystem — https://dl.acm.org/doi/pdf/10.1145/3631537
31. [s31] NVIDIA官方 · NVIDIA Story（中文） — https://images.nvidia.cn/aem-dam/zh_cn/Solutions/about-us/documents/NVIDIA-Story-zhCN.pdf
32. [s32] 电脑王阿达 · 黃仁勳 GTC Taipei 演講全紀錄 — https://www.koc.com.tw/archives/644529
33. [s33] 36氪 · 黄仁勋敞开心扉聊十大话题 — https://m.36kr.com/p/3741129232990473
34. [s34] GlobalCoachGroup · From Denny's Dishwasher to AI Architect — https://globalcoachgroup.com/from-dennys-dishwasher-to-ai-architect-the-unconventional-leadership-of-jensen-huang/
35. [s35] Stephen Witt · The Thinking Machine (Book)
36. [s36] Tae Kim (Barron's) · The Nvidia Way (Book)
37. [s37] stockstar.com · Jensen Huang: 英伟达持股与减持 — https://wap.stockstar.com/detail/IG2025060500008138
38. [s38] 豆瓣书评 / 澎湃新闻 · 黄仁勋：英伟达之芯 — https://book.douban.com/review/16388961/
39. [s39] NVIDIA官方 · NVIDIA Vera Rubin Official Page — https://www.nvidia.com/en-us/data-center/vera-rubin-nvl72/
40. [s40] Anthem Creation · Jensen Huang declares AGI achieved - Analysis — https://anthemcreation.com/en/artificial-intelligence/jensen-huang-declares-agi-achieved-analysis/

---
本档案机器版由亚洲封面人物官网自动生成：https://coverfigure.com ｜ 永久档案：https://coverfigure.com/acf/ACF-00-00160/geju ｜ 生成时间 Generated：2026-09-11 00:42